Achieving Six Sigma Printed Circuit Board Yields
by Improving Incoming Component Quality and
Using a PCBA Prioritization Algorithm
By
Daniel Jacob Davis
B.S.E Mechanical Engineering, The University of Michigan, 2001
Submitted to the MIT Sloan School of Management and the Mechanical Engineering Department
in Partial Fulfillment of the Requirements for the Degrees of
MASACH-SETS INSTRJTE
OF TECHNOLOGY
Master of Business Administration
(
AND
JUN 2 5 2008
Master of Science in Mechanical Engineering
LIBRARIES
In conjunction with the Leaders for Manufacturing Program at the
Massachusetts Institute of Technology
June 2008
C 2008 Massachus tts Institute of Technology. All rights reserved
Signature of Author
, .Certified by
flhepartment of Mechanical Engineering &
MIT Sloan School of Management
May 09, 2008
Certified by
David Hardt, Thesis Supervisor
Ralph E. and Eloise F. Cross Professor of Mechanical Engineering
Certified by
Professor of Statistics and "gement
Rq/ Welsch, Thesis Supervisor
Science and Engineering Systems
Accepted by
Lallit Anand, ~
ate Committee Chairman
Department of Mechanical Engineering
Accepted by
B
Debbie Berechman
Executive Director of MBA Program, MIT Sloan School of Management
This page has been intentionally left blank
Achieving Six Sigma Printed Circuit Board Yields
by Improving Incoming Component Quality and
Using a PCBA Prioritization Algorithm
By
Daniel Jacob Davis
Submitted to the MIT Sloan School of Management and the
Mechanical Engineering Department on May 09, 2008 in Partial Fulfillment of the
Requirements for the Degrees of Master of Business Administration and
Master of Science in Mechanical Engineering
ABSTRACT
Printed circuit board assemblies (PCBAs) are the backbone of the electronics industry. PCBA
technologies are keeping pace with Moore's Law and will soon enable the convergence of video,
voice, data, and mobility onto a single device. With the rapid advancements in product and
component technologies, manufacturing tests are being pushed to the limits as consumers are
demanding higher quality and more reliable electronics than ever before.
Cisco Systems, Inc. (Cisco) currently manufactures over one thousand different types of printed
circuit board assemblies (PCBAs) per quarter all over the world. Each PCBA in Cisco's
portfolio has an associated complexity to its design determined by the number of interconnects,
components, and other variables. PCBA manufacturing yields have historically been quite
variable. In order to remain competitive, there is an imminent need to attain Six Sigma PCBA
yields while controlling capital expenditures and innovating manufacturing test development and
execution. Recently, Cisco kicked off the Test Excellence initiative to improve overall PCBA
manufacturing yields and provided the backdrop to this work study.
This thesis provides a first step on the journey to attaining Six Sigma PCBA manufacturing
yields. Using Six Sigma techniques, two hypotheses are developed that will enable yield
improvements: (1) PCBA yields can be improved by optimizing component selection across the
product portfolio by analyzing component cost and quality levels, and (2) Using the Six Sigma
DMAIC (define-measure-analyze-improve-control) method and the TOPSIS (Technique for
Order Preferences by Similarity to Ideal Solutions) algorithm, PCBA yields will improve by
optimally prioritizing manufacturing resources on the most important PCBAs first. The two
analytical tools derived in this thesis will provide insights into how PCBA manufacturing yields
can be improved today while enabling future yield improvements to occur.
Thesis Supervisor: Dave Hardt
Title: Ralph E. and Eloise F. Cross Professor of Mechanical Engineering
Thesis Supervisor: Roy Welsch
Title: Professor of Statistics and Management Science and Engineering Systems
This page has been intentionally left blank
Acknowledgments
I am very fortunate to have worked with such great, smart people during my internship. I want
to thank Cisco for giving me the opportunity to work on such an interesting project. Mike
Lydon, my internship sponsor, and Gary Cooper, my supervisor were extremely instrumental in
making my internship both inspiring and rewarding. Mike and Gary gave my project full support
and visibility throughout Cisco. Without you both, my experience would not have been as
exciting, stimulating, challenging, or successful. I had a wonderful, truly invaluable experience.
During my internship, I had the honor to work with many other terrific people from Cisco. I
want to especially thank Vah Erdekian, Greg Jordan, Roger Bhikha, Bill Eklow, Zoe Conroy,
Erich Shaffer, Dave Towne, Hanbo Wang, Steve Nunamaker, Jim Leidigh, Paul Bennett, Raj
Saxena, Derrick Kidani, Sachin Kothawade, Leslie Averbeck, Hitesh Merchant, Deepak Pathak,
Ali Nouri, Peri Ryan, Charlotte Jackson-McCowan, and the rest of the Test Excellence team.
You all were extremely influential in helping me create and shape my ideas for this thesis.
The MIT Leaders for Manufacturing (LFM) community at Cisco was also very helpful and
supportive with my project. Jim Miller and Prentis Wilson, senior LFM leaders at Cisco, gave
my project full support. The LFM network allowed me to get in touch with the key stakeholders,
develop my ideas, and refine my solutions - thank you Miriam Park, Erik Stewart, Johnson Wu,
Chris Pandolfo, and Julie Go. I also want to thank Chris Richard, my LFM mentor, for his
advice, dedication, and support.
It was a pleasure to work with my two advisors: Professor Roy Welsch and Professor Dave
Hardt. Professor Welsch was fully engaged in my internship, helping zero it in on a specific
topic as he often called it "12 LFM internships in one." Professor Hardt was extremely
influential in my project and helped refine my approach while always offering keen insights.
Lastly, I want to give a huge thanks to Don Rosenfeld (aka "The Don"). I don't know how you
do it year over year, but I really appreciate your efforts. Without the support from my MIT
advisors and the MIT staff, I would have never been able to complete this work.
Lastly, I want to thank my very supportive family and especially, the most important person in
the world to me, my future wife, Julie Glick. (sorry I missed that weekend in NYC - I had to
write the thesis.) Julie is my inspiration! She is a woman with a tremendous personality that
challenges me in everything I do, including every sentence written in this thesis.
Thank you all for the support and advice in my life.
This page has been intentionally left blank
Table of Contents
Acknowledgments ........................................................................................................................................
5
Table of Contents ........................................................................................................................................
7
Table of Figures ..
.................................................................
10
Table of Tables ....................................................................................................................................
11
Table of Equations ..............................................................................................................................
12
1.
13
13
15
17
18
20
Chapter 1: PCBA Introduction, Cisco, and Motivation to Change .............................................
1.1. The Importance of Printed Circuit Board Assemblies .....................................
1.2. The Printed Circuit Board Assemblies Manufacturing Process .................................
1.3. Introduction to Cisco Systems, Inc ...................................... ......
................
1.4. Impetus for Change........................................................................................................
1.5. Printed Circuit Board Yields and Attaining a 6a Process...............................
1.6.
1.7.
Test Excellence Vision ........................................... ................................................. 21
My W ork-Study: One part Management, Two parts Engineering........................ . 22
1.7.1.
One Part Management Opportunity ................................................
1.7.2.
Two Parts Engineering Opportunity ................................................
1.8. Two Hypotheses.............................................................................................................
1.8.1.1.
Hypothesis 1..................................................................................................
1.8.1.2.
Hypothesis 2.................
22
22
23
23
........................................................................... 23
1.9 . Summ ary ........................................................................................................................
23
1.10. Thesis Approach & Structure...................................... ............................................. 24
2.
Chapter 2: Research Methodology............................................................................................
27
2.1. Assess the Current Situation ........................................................... ......................... 28
2.2. Determine the Future State ...........................................................................................
28
2.3. Develop a Hypothesis .................................................
............................................. 29
2.3.1.
H ypothesis 1......................................................... ............................................ 29
2.3.2.
Hypothesis 2...........................................................................................................30
2.4. Research and Literature Review .........................................................................
31
2.4.1.
Component Quality Optimization Model ........................................... 32
2.4.2.
The PCBA Prioritization Algorithm ................................................ 35
2.5.
Summary ...................................................
37
3. Chapter 3: Hypothesis 1 Results - The Component Quality Yield Optimization Model to
achieve a 6a Yield Process .....................................................................................................................
39
3.1.
Introduction to PCBA Yield Calculation .........................................
............. 46
3.2. Probabilistic Yield Model based on Components Defect Rates ................................. 47
3.3. The Component Quality Yield Optimization Model ............................................... 49
3.3.1.
Determine the 60 Budgets for Each Component ....................................... 51
3.3.2.
Calculated the Cost of Poor Quality (EI) for each PCBA................................
57
3.3.3.
Calculated the Cost of Investm ent ( 2) ......................... .................................... 57
3.3.4.
Set up the Nonlinear Program and Optimize ......................................
.... 59
3.4. Results and Application of the Component Quality Yield Optimization Model ........ 62
3.5. Limitations of the Component Quality Yield Optimization Model ............................ 70
3.6. Future R esearch ................................................ ....................................................... 7 1
3.7. Summary of the Component Yield Optimization Model....................................
72
4.
Chapter 4: Hypothesis 2 Results - The PCBA Prioritization Algorithm using 6 ..................
4.1.
D efine: The Problem Statem ent........................................
................................
4.2. Measure: The PCBA Prioritization Algorithm .....................................
.........
4.2.1.
Scoring each PCBA Based on Manufacturing Data .............................................
4.2.2.
The Revenue & Demand Index Score (RDI) .....................................
.....
4.2.2.1.
RDI Factor 1 & 2 - Demand Forecast (a ) and Next Year (3) ...................
75
75
77
78
82
82
4.2.2.2.
RDI Factor 3 y , - Ratio of Volume to Cost .......................................
82
4.2.2.3.
Putting the Revenue & Demand Index together ....................................
83
4.2.3.
The Quality Index Score (QI) ..................................... ....................... 85
4.2.3.1.
QI Factor 1 - 6 , The 60 OR Perfect Yield Delta .................................... 85
4.2.3.2.
QI Factor 2 - E , The 6o OR Perfect Yield Delta 13 week Trend ................ 86
4.2.3.3.
QI Factor 3 - ý, The Cost of Poor Quality (COPQ) ................................. 86
4.2.3.4.
QI Factor 4 - rl , Eliminating the Waste in the System ............................. 86
4.2.3.5.
QI Factor 5 - 0 , The Ratio of COPQ to Waste..................................
87
4.2.3.6.
Other Costs to Consider ...................
87
4.2.3.7.
Putting the Quality Index together .......................................
...............
...... ... 87
4.2.4.
The Customer & Management Index Score (CMI) ........................................
4.2.4.1.
CMI Factor 1 - t, The Customer Experience............................
......
88
89
4.2.4.2.
CMI Factor 2 - K , The Market Importance .......................................
4.2.4.3.
CMI Factor 3 - k, Quality Engineers' Expected Performance .................. 90
4.2.4.4.
Putting the Management and Customer Index together............................ 90
89
4.3. Analyze: Putting the PCBA Total Index Together using TOPSIS ............................. 92
4.3.1.
TOPSIS-Step 1: Obtain PCBA Raw Scores & Determine the Ideal States..........92
4.3.2.
TOPSIS Step 2: Normalize the Raw Scores ......................................
..... 94
4.3.3.
TOPSIS Step 3: Weight the Normalized Scores .....................................
.... 95
4.3.4.
TOPSIS Step 4: Determine the Priority Index Based on the Ideal States...........96
4.3.5.
TOPSIS Step 5: Display the Index Scores .................... .................................... 98
4.3.6.
Apply General Priority Rules to the Total PCBA Index........................................99
4.3.7.
Generate Final Priority Rankings on the Total PCBA Index............................ 101
4.3.8.
Validate using 27 Extreme Corner Case Scenarios ............................................. 104
4.3.9.
Summary of the PCBA Prioritization Algorithm............................
105
4.4. Improve: The PCBA Prioritization Algorithm and Overall PCBA Yields .................. 106
4.4.1.
Improvements using Different Weighting Scenarios.............................106
4.4.2.
Improve PCBA Yields Today .............................................................................. 10
110
4.5. Control: Monitor PCBA Metrics over Time ..........................................................
111
4.6. Summary..........................................................................................................................
5.
Chapter 5: Organizational Design and Implementing Change .......................................
5.1.
5.2.
5.3.
5.4.
5.5.
5.6.
C om pany Intro .............................................................................................................
The Strategic Lens ........................................
The Political Lens ........................................
The Cultural Lens ........................................
Combining the Three Lenses .....................................
Sum m ary ......................................................................................................................
6. Chapter 6: Conclusion
........................................................
Bibliography ..............................................................................................
..................................
113
113
115
122
126
134
135
137
139
Table of Figures
Figure 1: General PCBA Manufacturing Process Flow..................................
............. 15
Figure 2: Recent Cisco Acquisitions 2005 - 2007 .........................................
.............. 17
Figure 3: Total Cost Curve and Associated Yield Fallout for Different Component Selections . 68
Figure 4: PCBA Yield - Cost Efficient Frontier .............................................................
68
Figure 5: Three Steps to Determine the Final PCBA Priority List .....................................
78
Figure 6: PCBA Prioritization Inputs .....................................................
............................... 81
Figure 7: C alculating the R DI ................................................................................................
Figure 8: C alculating the QI .........................................
84
........... ........................................ 88
Figure 9: C alculating the C M I ...................... .....................
................................ . 91
Figure 10: General Priority Rules ...........................................................................................
Figure 11: Four General Rules Tie into the Final Priority List...........................................
99
101
Table of Tables
Table 1: Select List of Components Used in Printed Circuit Board Assemblies...................... 39
Table 2: Costs Associated with Component Quality .......................................
........... 41
Table 3: Component A's Quality Level and Cost for Each Quality Standard.......................... 43
Table 4: Decision Variables to Determine 60 Component dpmo Budget ................................ 52
Table 5: Constraints for Determining 60 Component Budgets by Minimizing A2 ................... 53
Table 6: 6c dpmo Budgets so Every PCBA Achieves a 60 Yield...................................
54
Table 7: Summary of "As-Is" dpmo Goals for 5 PCBA's Component Yield Fallout............... 55
Table 8: Summary of the 60 dpmo Budgets for 5 PCBA's Component Yield Fallout .........
55
Table 9: Different Quality (dpmo) Levels for Each Component ......................................
56
Table 10: Cost of Investment,'
2a
Cost of Improving the Component Quality ...................... 58
Table 11: Cost of Investment,'
2b
Cost of the Component ....................................
Table 12: Weighting Scenarios for the Objective Function................................
......
59
.......... 60
Table 13: Decision Variables for Optimization Model...................................
............ 61
Table 14: Scenario I's Component Selection Results .......................................
........... 62
Table 15: Summary of Scenario I's Optimization Results .....................................
........ 63
Table 16: Scenario I's Component Yield Fallout Percentage by PCBA ................................... 63
Table 17: Scenario II's Component Selection Results ......................................
........... 64
Table 18: Summary of Scenario II's Optimization Results................................
.......... 65
Table 19: Scenario II's Component Yield Fallout Percentage by PCBA ................................. 65
Table 20: Summary of the Optimization Results for Scenarios I - V........................................ 66
Table 21: Example of Manually Selecting All 60 Components .......................................
Table 22: Summary of the Results for Scenarios VI - X ......................................
66
......... 67
Table 23: Important Manufacturing Factors Used to Calculate Each Index ............................. 80
Table 24 : Ideal States ............................................... ............ .................................................
93
Table 25: Ten PCBA Normalized Scores for Each Index ......................................
95
Table 26: Ten PCBA Weighted Scores for Each Index.................................
........
............... 96
Table 27: Priority Index for Each Index .................................................................................. 97
Table 28: Index Scores Based on the Priority Index...............................
................... 98
Table 29: The PCBA Total Index with Comments ...............................
102
T ab le 30 : D OE M atrix ............................................................................
................................
104
Table 31: DOE for 27 PCBAs ........................................
105
Table 32: 20 Randomized PCBA samples.....................................
107
Table 33: Different Weighting Scenarios .....................................
108
Table 34: Different Weightings Scenarios Applied to 20 Randomly Sampled PCBAs......... 108
Table 35: Weightings Scenarios Applied to 18 Revenue Generating DOE Extreme Cases ...... 109
Table 36: Test Engineer Responsibilities within Different Manufacturing Departments ....... 118
Table of Equations
Equation 1: Total Com ponent Cost.........................................................
................................ 41
Equation 2: Functional Yield Equation for any PCBA......................
............ 47
Equation 3: Probability of PCBA Not Failing due to the ith Component ................................. 48
Equation 4: Calculating the Component Yield Fallout% using a Probabilistic Model ................ 49
Equation 5: PCBA TCC Equation ................................................. .... .................................... 49
Equation 6: C ost of Investm ent ............................ .. .............
...................
...................... 50
Equation 7: Indices as a Function of the Factors ................................ .... ........................... ... 80
Equation 8: Normalizing Each Factor's Score ......................
Equation 9: Calculating D ..............................
....
........ ..
................... 94
.........................
Equation 10: C alculating D+ .....................................................................
Equation 11: Calculating the Priority Index ...........................................
96
.........................
97
.................
97
1. Chapter 1: PCBA Introduction, Cisco, and Motivation to Change
1.1.
The Importance of Printed Circuit Board Assemblies
Imagine a world without electronics. No computers. No cell phones. No databases or social
networks. No video games. Life would be very different. No e-mail. No e-Commerce. No
Internet. No iPod. Companies like Google, Apple, Cisco, Intel, Microsoft, IBM, and Facebook
would not exist as they do today. Life would be extremely different.
Today, our society greatly depends on electronics, and electronics greatly depend on printed
circuit board technologies. As customers demand higher quality, more reliable products,
manufacturing printed circuit boards at higher yield and quality levels will be much more
critical. The invention of the microprocessor and the ability to compute and transfer data
circuitry gave birth to the information age and changed our world forever. The technology made
it possible to deliver information amongst people, businesses, and governments at speeds faster
than ever before. As Moore's Law drove better performing technologies at lower costs, new
products took advantage and innovations rapidly occurred. People started to communicate in
new ways: through bits and bytes, l's and O's, fiber optics, and satellites. Imagine how many
industries, jobs, and markets electronic innovations have created. Software, hardware,
networking, security, internet, search, mobile - and we are just getting started. Prices will
continue to drop and future price decreases will stimulate future demand, and thus, increase
innovation (Nystedt, 2008).
Printed circuit board assemblies (PCBAs) hold the fundamental circuitry to transmit analog,
digital, or optical information. PCBAs enable information to be extracted, processed, analyzed,
synthesized, displayed, and transferred amongst and between electronic devices such as
computers, cellular phones, mobile devices, modems, routers, internet switches, and digital
cameras, allowing the world to stay connected while enabling innovations to occur faster than
ever. PCBAs are the platforms that tie all these technologies together and are the central nervous
system of electronics enabling the information age to rapidly expand and the world to flatten
(Freidman, 2006). PCBAs are the fabric that has allowed the information age to exist and
expand, the digital revolution to occur and evolve, and Web2.0 technologies to be created and
virally proliferate.
With more computing power available to more of the world, people, businesses, and
governments have been able to connect more easily while productivity has increased.' As PCBA
technologies continue to keep pace with Moore's Law, PCBAs have become more complex and
more powerful while decreasing in size and cost. This has allowed many new technologies to
start taking full advantage of current PCBA innovations - enabling the convergence of video,
voice, data, and mobility into one single electronic device.
PCBAs are designed and manufactured for a variety of technologies and applications used
around the world, and the electronics and information industries keep demanding better, higher
quality products that are faster, cheaper, smaller, and greener. As products become smaller and
demand more functionality, so to must the PCBAs. PCBAs evolve ahead of the product enabling
new product innovations to occur. PCBAs continue to pack more technologies into a smaller and
smaller area. As a result, PCBA complexity 2 has dramatically increased and, therefore, so has
the number of defect opportunities (Oresjo, 2003). PCBA quality performance has been
increasingly more important as the circuitry has become more complex requiring more
components and solder joints than ever before. To remain competitive, PCBA manufacturing
costs must remain low. Therefore, PCBA manufacturers must quickly learn how to produce
more complex PCBAs at higher yields and lower costs (Tong, Tsung, & Yen, 2004).
Manufacturing these ever changing, increasingly complex PCBAs becomes more and more
challenging as more sophisticated circuitry is introduced while the number of components per
PCBA increases. Firms are constantly introducing new products and creating new markets with
innovative products. Customers continue to demand higher quality products that are more
reliable. Therefore, to remain competitive, PCBA manufacturing firms must produce a high
yielding PCBA that translates into a high quality, highly reliable product. Utilizing the Six
1Fine, Charles. MIT Lecture Notes. Spring
2008. Course 15.769: Operations Management.
2 On a very simple level, complexity is measured by the number of components and solder joints per PCBA.
Sigma (6c) philosophy of achieving 3.4 defects per million opportunities (dpmo), techniques
such as DMAIC (define-measure-analyze-improve-control) will assist in improving PCBA yields
(Bafiuelas & Antony, 2005, p. 251). The research and ideas for this thesis stem from a seven
month work-study at Cisco Systems, Inc (Cisco) in San Jose, California. This thesis addresses
methods that will help bring all PCBA yields one step closer to meeting the 60 yield goal. We
will discuss Cisco and achieving a 6c process in more detail in sections 1.3 to 1.5, but first, let's
quickly review the PCBA manufacturing process.
1.2.
The Printed Circuit Board Assemblies Manufacturing Process
PCBAs are designed and manufactured all around the globe. For instance, Cisco Systems is a
networking company that designs, manufactures, and packages solutions that use multiple
combinations of different types of PCBAs. PCBAs are manufactured and tested in several
process steps simplified below in Figure 1.
Figure 1: General PCBA Manufacturing Process Flow
Place
Manufacture
Components
PCB circuitryon
on PCB
PCB
Solder
Components
/ to
to PC
PCB
Test PCBA
First, on a non-conductive substrate, the printed circuit board (PCB) circuitry is manufactured,
establishing the highways for electrons to travel between components. Next, components are
placed in the correct position and orientation on the PCB. After correct placement, the
components are securely soldered to the board to ensure they have proper connection to the
circuitry. This printed circuit board assembly (PCBA) is then subjected to multiple tests.
Several types of tests exist in manufacturing to determine the health of the manufacturing
process. These tests can be classified as structuraltests orfunctional tests.
These tests will determine if the PCBAs are made correctly, verify if the PCBA is properly
assembled, both structurally and functionally, testing for as many opportunities that may cause
the PCBA to fail. Once all the components are securely placed on the PCBA, several structural
tests run to determine if the components were placed in the proper location and in the proper
orientation. Structural tests also check to make sure solder paste is applied correctly and that
connections are mechanically sound. Structural tests look for potential short or open circuits that
would cause the PCBA to fail and can also include stressing the board at extreme thermal and
vibration conditions.
The functionaltests verify that the PCBA works as it was designed to function. The number of
and the type of functional tests vary by product but generally run multiple diagnostics that
electronically exercise all the components on the PCBA at various temperatures and voltages to
ensure the PCBA remains functional at its specifications.
Testing is a non-value added step because it does not change the product in any way and
increases the overall cycle time of the product. However, each test acts as an insurance policy
that allows defective parts to be screened out of the process, and thus, reduce the amount of
lower quality product that could be delivered to customers. Additionally, the overall health of
the manufacturing process is determined by the products' yields, which are calculated from test
data. Although tests are non-value added operations, the information extracted from the tests
provides visibility into the manufacturing process and allows for improvements to be made. As
more test information allows product yields to improve, tests can then be optimized (Oresjo,
2003) and even eliminated over the life cycle of the product.
As the electronics industry promises to delivery better quality products, quality excellence has
come to the forefront of many PCBA companies. In the past, being the first mover and focusing
on a time to market metric often competed with the product's quality. At Cisco, the company is
engaged in a 60 effort to create a culture that prioritizes quality first. Through Manufacturing
and Test Excellence Initiatives, Cisco strives to manufacture every PCBA in its portfolio at its 60
goal. Utilizing the information from current tests will allow the company to make certain
business decisions that will ultimately improve quality levels and reduce costs.
1.3.
Introduction to Cisco Systems, Inc.
Cisco currently manufactures over 1,000 different types of printed circuit board assemblies per
quarter all over the world. (See Chapter 5 for a much more in-depth organizational analysis of
Cisco). Each printed circuit board assembly, or PCBA, in Cisco's portfolio has an associated
complexity to its design determined by the number of solder joints (known as interconnects),
components, and other variables. Low complexity boards may have 200 interconnects and 10
components while high complexity boards may have over 40,000 interconnects and 5,000
components. Cisco's Test Excellence Initiative was kicked off in the summer of 2007 and has
been tasked to implement the optimal tests process for each PCBA while improving overall
PCBA yields for the entire Cisco product portfolio.
Cisco was founded in 1982 and to date has made 125 acquisitions with the most recent seen in
Figure 2. Currently, the company is 100% outsourced with strategic in-house manufacturing
centers. Cisco offers a wide variety of product complexity, from low end IP phones to very high
end routers. Over 250,000 orders are processed every quarter on 196 active product families
with over 23,000 product identification numbers. Additionally, there are 600 suppliers with
50,000 purchased part numbers.
Figure 2: Recent Cisco Acquisitions 2005 - 2007
Category
,Il
Consumer
O Video 67%
SApplicatlor
27%
Secuiaty7%
Topspl
'
Data Center 4%
Alrespl
IronF
T
Mobility4%
" Components <%
1_
_j
.....
__ __93% of these nvestments
cat nw compettors
6%are acorreeve change inexisdng strtegy
<1%are foundation Investmentfor core business
-
Management
Softwarem 1%
1 Services 0.1%
Source:Gartner(January2008)
As Cisco continues to grow, the supply chain complexity and scope of manufacturing grows too.
However, to remain competitive, the old one-size-fits-all test philosophy that has helped make
Cisco such as success must change. In order to remain competitive and low cost, an optimal
testing strategy must be implemented. But is it worth it to change for the sake of change, or are
there external driving factors that require change to occur?
1.4.
Impetus for Change
Globalization is making the world flatter. Consumers and businesses want information ondemand, in real time, and in the palm of their hand. As the electronics industry continues to
diversify into the consumer and commercial segments while focusing on the convergence of
voice, video, data, and mobility, innovative products are pushing the limits of test requirements.
Newer tests need to be developed and deployed to ensure high quality products meet customer
expectations. At the same time, the industry aspires to attain 6Y quality and production
flexibility while controlling capital expenditure. To sustain a competitive advantage,
manufacturing companies need to flawlessly delivery high quality products at the highest
possible yield, utilizing an appropriate test suite that optimizes risk, affordability, and test
coverage.
The current test strategy was intended to get Cisco to $40B in revenue by 2004, and it was very
successful in its mission. However to remain competitive, Cisco believes that it is imperative to
change the test strategy. As the company grew by acquisition, the test strategy did too. This 11
year old test strategy focuses on testing for escapes rather than designing in quality. It does not
significantly differentiate by product type or by product lifecycle stage. As Cisco moved to a
100% outsourced manufacturing strategy, the test strategy did not optimize across its network of
contract manufactures and vendors.
The business and technology environment is right for change. Realizing that all competitive
advantage is temporary (Fine, 1998), Cisco knows their test strategy must change. Cisco has a
mission to continuously improve and drive manufacturing excellence. Thus, Cisco is dedicated to
fostering an environment to achieve world class operations.
To understand and analyze any manufacturing process, tests must be completed. However, what
is the right amount of testing that should be done to ensure the process is capable, efficient, and
economical? Implementing tests in a manufacturing process incurs both cost and time, both of
which companies are quickly trying to reduce. Market pressures are driving prices down while
customers are expecting their product to arrive quicker than ever before with higher quality and
reliability. Therefore, it is imperative companies understand their test operations.
On one hand, if no money was spent on testing the process, products would have very low lead
times. However, inevitable defects in the manufacturing process would be passed downstream
and eventually to the field. These defects could result in failed parts, product recalls, and
ultimately lead to poor customer satisfaction and potential loss of market share. So, even though
no cost was incurred with the physical tests at the manufacturing site, several costs will be
incurred because of manufacturing defects that translate to field failures, product recalls, and
customer dissatisfaction, otherwise known as the cost ofpoor quality (COPQ).
On the other hand, if a company were to spend unlimited funds on manufacturing tests, product
costs and lead times would skyrocket. This, in turn, would lead customers to choose
competitor's products with lower prices and quicker lead times. Tests are a very important part
of any manufacturing process. The data extracted from manufacturing tests help management
understand the health of the entire process and detail critical indicators such as lead times,
product yields, component defect rates, overall factory yields, and inventory levels. With cost
pressures and emerging markets, it is very important to improve and sustain world class factory
yields as defined by 6a yield performance.
So what is the optimal amount of test to support a 6a yield process at the lowest cost possible?
This is exactly the question Cisco faces as the company embarks upon an initiative to modernize
its testing philosophy. In the past, Cisco applied a one-size fits all test approach: Test
everything the same. The management team realized that this cookie cutter testing philosophy is
not scalable and that changing the testing strategy is imperative for the company to sustain its
competitive advantage. Cisco's launch of the Test Excellence Initiative will deliver these answers
through many different projects. Achieving a 6a process will not happen overnight or with one
project. No, this will be company wide, collaborative effort. This thesis focuses on two issues
of the many that will aid in attaining Sigma yields: (1) incoming components and (2) optimizing
which PCBAs to improve first.
1.5.
Printed Circuit Board Yields and Attaining a 6a Process
Tests are used in manufacturing to measure the health of the manufacturing process. These
manufacturing tests calculate PCBA yields. Historically, Cisco yields were a "straight yield"
calculation, based on the number of bad products divided by total number of products produced.
However, many different types of PCBAs can be manufactured on the same line and comparing
yields of different complexity types of PCBAs is like comparing apples to oranges.
PCBAs range in complexity. Higher complexity products are used for more complex
applications. Typically, the higher the number of component and solder joints per PCBA, the
more complex the PCBA. For example, PCBA-1 is a low complexity board with 200 solder
joints, 10 components, and an annual volume of 1,000,000 units. On the other hand, PCBA-2 is
a high complexity board with over 40,000 solder joints, more than 5,000 components, and an
annual volume of 100 units. Both PCBA-1 and PCBA-2 could be manufactured on the same
line, so comparing their "straight yield" calculation does not make sense.
Today, Cisco normalizes PCBA yields by taking many different variables into account when
making the final yield calculation. Namely, solder joints and components make up a large
percentage of potential opportunities for yield fallout, but there are many other variables
responsible too. This thesis focuses on understanding how component quality will affect overall
PCBA yields. Improving component quality will assist in allowing Cisco to achieve 60
manufacturing process across every product it manufactures.
A 60 process is based on statistics and is defined as 3.4 defects per one million opportunities
(dpmo), often referred to as 3.4 dpmo. Products that yield worse than 3.4 dpmo produce
unnecessary waste in the manufacturing system. The waste comes in many different forms such
as products that are scrapped, failures at the manufacturing site, or returns by the customer and
should not be passed along in the system (Ohno, 1988). For every type of waste event, there is
an incurred cost, and the cost increases as the product moves downstream in the manufacturing
process. Thus, a customer return is the highest cost, the most detrimental type of waste.
To eliminate the waste in the system, Cisco must invest money to improve their capabilities
(Repenning & Sterman, 2001). Spending money earlier in product development and
implementing optimal tests in the manufacturing process will create higher quality products that
will achieve the targeted 3.4 dpmo. This goal will ultimately eliminate unnecessary waste in the
system and allow high quality products to be delivered to customers.
Achieving a 6a process is easier said than done, especially as PCBA complexity increases. First,
if a PCBA is to achieve a 6a process, the PCBA must have 3.4 dpmo or fewer. What is a PCBA
opportunity? If we assume that any one item on a PCBA can cause a failure, then every item is
one opportunity.
Imagine two examples, 1) a PCBA with 100 opportunities and 2) a PCBA with 70,000
opportunities. In the first example, one failure would make this specific board have 1 defect in
100 opportunities, or 10,000 dpmo. Thus, if one million of these boards were made, 10,000
would be thrown away. To achieve a 6a process, example 1 should only have 0.00034 defects
per 100 opportunities, or an improvement of nearly four orders of magnitude. In the second
example, one failure would make this specific board have 1 defect in 70,000 opportunities, or
14.29 dpmo. To achieve a 60 process, example 2 should only have 0.238 defects per 100
opportunities, or an improvement of one order of magnitude.
These examples are not unrealistic and represent the challenges Cisco faces in becoming a world
class PCBA manufacturer. However, improving the design and manufacturing process in the
PCBA industry to be capable of achieving a 60 process is going to take money, time, and a
cultural paradigm shift.
1.6.
Test Excellence Vision
The Test Excellence Initiative will drive the 6a culture change. Cisco has embarked on a
journey to ensure future success through this company wide, collaborative initiative. Test
Excellence is meant to provide a venue to foster an environment to become a world class, 60
manufacturing firm. Utilizing the current business environment and the impetus for change, Test
Excellence will create a paradigm shift within the company and ultimately a new way of thinking
across Cisco. Test Excellence is intended to become part of Cisco's DNA, and ultimately, Cisco
envisions that every PCBA it manufactures will achieve its 6o yield goal while optimizing risk,
affordability, and test coverage.
1.7.
My Work-Study: One part Management, Two parts Engineering
The Test Excellence initiative will modernize the current test strategy preparing Cisco's test
infrastructure to capitalize on the anticipated growth for Web2.0 and beyond. My work-study
focused on designing a world class test strategy and governance model to enable an agile,
aligned, and adaptive supply chain.
1.7.1. One Part Management Opportunity
At the crux of the internship, I managed the entire Test Excellence Initiative and participated
on the four main sub-teams. The initiative fostered communication channels within the
company to create new ideas and help with the paradigm shift to become world class in
manufacturing high quality, 6o PCBA products. The team was composed of four man Cisco
divisions: Cisco Design Organization, Product Operations, Manufacturing Operations, and
Technology and Quality.
1.7.2. Two Parts Engineering Opportunity
During my participation on the core and sub-teams, I found opportunities for improvement,
specifically, in yield management. First, it is not well understood how different components
affected the overall yield of PCBAs. What quality level is needed for each component to
achieve the overall 6o yield goal? Second, I found that yields were managed at local levels
rather than as a portfolio of products. Therefore, it was difficult to find the best and worst
performing PCBAs across the different manufacturing sites. These two engineering
challenges help form my two hypotheses for this thesis.
In short, the end goal of Test Excellence is all PCBAs yielding at the 60 goals with an optimal
test plan that evolves over the product's lifecycle. On the journey to this ideal state, it is
imperative to first understand the effect of components quality on overall PCBA yields, as well
as understand the current state of product yields across the company's manufacturing footprint.
1.8.
Two Hypotheses
Before the company can optimize manufacturing tests plans across all their products and
improve yields, it is necessary to first look at how component quality affects yields. With this
understanding, tests can then be added, eliminated, or changed on a product by product basis to
align the tests to the market demands. Additionally, as the test plans evolve over the product's
lifecycle, managing test yields for the entire product base is needed to help better optimize test
development in the future. Therefore, this thesis details two hypotheses that are inputs into
optimizing the test over the lifecycle.
1.8.1.1.
Hypothesis 1
Hypothesis 1: The Component Quality Yield Model Hypothesis
Yields and costs can be optimizedfor an entireportfolio of PCBA products by selecting the
appropriatecomponents based on component quality and cost specifications.
1.8.1.2.
Hypothesis 2
Hypothesis 2: The PCBA Prioritization Algorithm
Overall PCBA yields will improve by optimally allocatingmanufacturingresources to a
holistically prioritizedlist of PCBAs across the entire portfolio.
1.9.
Summary
In this introductory chapter, we looked at why PCBAs are so very important to our society and
how PCBAs are manufactured. We took a closer look at Cisco and noted the business
environment supports the change to achieve a world class, 60 PCBA manufacturing process.
Supporting the need to change, Cisco has launched an internal Test Excellence initiative to
implement a new test strategy to ultimately improve product yields while lower manufacturing
costs.
While working on the Test Excellence initiative during my seven month work-study at Cisco, I
uncovered specific yield issues that allowed the formation of two hypotheses to improve overall
PCBA yields. Vendor component quality effects yield fallout, but the magnitude of yield fallout
as a function of the number and type of components is not fully understood. Additionally, Cisco
does not prioritize the PCBAs across the company's portfolio but rather allows over 110
specialized local teams to resolve yield issues with very little or no best practice sharing among
teams. This thesis further details how component quality affects each PCBA's yield while also
deriving a prioritization algorithm for the entire product portfolio to help local teams focus
attention on the most important PCBAs first. Finally, this thesis will analyze Cisco's
organizational design to understand how the firm will react to changes driven by Test
Excellence.
1.10. Thesis Approach & Structure
This thesis is divided into several chapters. Research used for this thesis involves industry
experts from Cisco, a thorough review of academic and industry literature, and guidance from
my advisors. As Chapter 1 discussed, this work-study was divided into two separate threads that
I worked on concurrently. The first thread involved the participation the Test Excellence
Initiative, a significant change management program to revamp the entire test strategy at Cisco.
The second thread involved more engineering work that involved identifying certain yield issues
and recommending a solution, both of which stem from the two hypotheses.
Chapter2 describes the research methodology for this thesis and how the hypotheses were
developed. This chapter also contains a literature review for both hypotheses. The work
performed at Cisco was used as a backdrop for the management and engineering thesis study.
Chapter3 details the analysis of the first hypothesis, the effect of component quality on overall
PCBA yields. This chapter further investigates how Cisco can use a nonlinear program to make
better management decisions when determining which components to use based on cost and
quality. Chapter 3 ends with recommendations for future research in this particular area.
Next, Chapter 4 details the analysis of the second hypothesis, an algorithm to prioritize
manufacturing resources to resolve yield issues on printed circuit board assemblies with the
highest return on investment. The tool allows management to holistically analyze its portfolio of
products based on a ranking system which will help optimize how it utilizes its manufacturing
resources to resolve the most important yield issues first. The ranking system is based on three
key inputs: (1) Revenue & Demand data, (2) Quality data, and (3) Customer & Management
data. Chapter 4 ends with recommendations for future research in this particular area.
Chapter 5 further discusses Cisco's organizational design. This chapter looks at Cisco from
three different perspectives: (1) The Structural Lens, (2) The Political Lens, and (3) The Cultural
Lens. It then hones in on how the organization is ready to adopt the tools from this thesis.
Finally, Chapter 6 concludes the thesis with a summary of the Component Quality Yield Model,
the PCBA Prioritization Algorithm, and Organizational Design analysis.
This page has been intentionallyleft blank
2. Chapter 2: Research Methodology
The two hypotheses for this thesis were developed as Cisco embarked upon a major
change initiative to create a new test strategy. Cisco's original test strategy was very
successful in enabling the company to achieve its current goals; however, in order to
remain competitive, the company's leaders realized the current test philosophy needed
change. One aspect of the new test strategy demanded that all PCBA yields meet their 6a
yield goals. Therefore, finding solutions to improve overall PCBA yields became a main
priority of the overall change initiative. To enable 6a yield goals to be met on every
PCBA manufactured requires company wide support and multiple solutions.
Consequently, this thesis is one of many first steps on the journey to achieving 60 yield
goals for every PCBA, and the results from these two hypotheses are just pieces of a
bigger puzzle.
The first hypothesis studies the effect of component selection on overall PCBA yields.
The second hypothesis investigates the optimal way to prioritize manufacturing resources
on many PCBAs to quickly improve yield issues. The two hypotheses lead to working
solutions that will enable Cisco to step closer to achieving a 6a process for all PCBAs.
Additionally, utilizing these solutions will enable future improvements to occur down the
road.
Developing the hypotheses required full participation in the change initiative and then a
detailed investigation of current problems. As a result, the first half of the work study
focused on implementing change within the organization while the second half
concentrated on solving more specific engineering problems. Therefore, before diving
into the two hypotheses, the following sections will discuss the methodology behind the
change management initiative.
At the beginning of the work study, Cisco kicked off the Test Excellence initiative to
revamp its current test strategy. The Test Excellence's change management process
enabled multiple meetings with current employees to collect data, understand the current
situation, and develop, test, and validate specific hypotheses. Finally, from many
potential hypotheses, two hypotheses were studied in greater detail. These hypotheses
were created, developed, and implemented through the following steps:
2.1.
*
Assess the current situation
*
Determine the future state
*
Develop a hypothesis
*
Research
Assess the Current Situation
To fully understand the problems associated with the current test strategy, multiple
interviews were conducted across the entire organization. The interviews were intended
to find major problems and improvement areas while understanding how the underlying
organizational strategy, culture, and political playing fields shaped the current situation.
Chapter 5 further examines the company based on the strategic, cultural, and political
three lens analysis (Carroll, 2006). After 153 interviews, a small team composed of
about fifteen individuals from across the organization worked to create a new test
strategy during several full day working sessions. The new test strategy's goal was to
guide the organization to collaborate across divisions while developing optimal solutions
for the major problems which currently existing with today's strategy. Several key
improvement areas surfaced during these all day working sessions, one of which was
improving overall PCBA yields.
2.2.
Determine the Future State
The new test strategy was deployed to the organization through the Test Excellence
change initiative. The Test Excellence initiative consisted of 50 people from across the
company, fostering an environment ideal for collaboration and change. In order to
improve PCBA yields, the team focused on defining the future state for all PCBA yields.
With the help from Cisco's leadership team, the future state would be to achieve the 60
yields goal for every PCBA.3 Additionally, every PCBA should be sustained at its 60
yield goal over its entire product lifecycle. The 60 yield goal allows Cisco to measure its
progress throughout the Test Excellence initiative as well as compare different business
units and manufacturing sites within the organization. Eventually, the process will enable
the company to share best practices and become a world class PCBA manufacturer.
Furthermore, achieving 60 yield goals will eliminate waste in the system while
improving product quality and customer satisfaction.
Achieving 60 yield goals on every PCBA is quite challenging since many variables
contribute to the overall yield of a particular PCBA. Therefore many solutions could
exist. Thus, a specific yield subteam formed to brainstorm solutions. Further detailed
interviews were conducted with key stakeholders that concentrated on how yields could
be improved. These interviews discussed several potential hypotheses in further detail
and then determined which solutions would be worth pursing. The results from the
interviews became the foundation for developing the two hypotheses for this thesis.
2.3.
Develop a Hypothesis
Improving overall PCBA yields is a current challenge and many potential solutions to the
problem exist. From PCBA designs to component selections to manufacturing processes,
many different variables affected the overall PCBA yield. After interviewing key
stakeholders and working on several teams with PCBA yield experts, two main themes
developed that suggested next steps for a detailed investigation. Thus, these topics were
the foundation for the two hypotheses in this thesis.
2.3.1. Hypothesis 1
Improving yields is a significant part of becoming a world class manufacturing firm.
With so many factors contributing to yield fallout, this thesis focuses on analyzing how
component selection impacts overall yields. Managers questioned what component
3 Recall from section 1.5 that the 6a goal will be different for every PCBA.
quality levels were needed to achieve a 60 process. Thus, understanding how
components affect overall PCBA yields became a major concern and the basis for
hypothesis 1.
Hypothesis 1: The Component Quality Yield Model Hypothesis
Yields and costs can be optimizedfor an entire portfolio ofPCBA products by selecting
the appropriatecomponents based on component quality and cost specifications.
A probabilistic model was built to predict yield fallout caused by component defect rates.
Using the bill of materials for several PCBAs, the number and type of components were
entered into the model. Several categories of component quality levels were defined,
such as the Gold, Silver, and Bronze standards. These different quality standards
represent the different choices that the firm can use for each component. Then, working
with a component quality expert, incremental costs to achieve these different component
quality levels were derived.4 Finally a nonlinear program optimizes overall PCBA yields
based on incoming component quality and cost variables. The model and results are
further discussed in Chapter 3.
2.3.2. Hypothesis 2
At the same time, PCBAs were not holistically prioritized across the entire product
portfolio but rather prioritized differently at each local site. For example, out of over
1000 PCBAs, it was not well understood which PCBAs needed immediate attention to
improve the current yield, or, in other words, which PCBAs had the highest return on
investment for yield improvements.
Because of manufacturing resource constraints, not every PCBA can be worked on
simultaneously. Manufacturing resources would be assigned to PCBAs with yield issues,
but the process was contained at the local manufacturing site level, completed in an ad-
4All
data in this thesis is disguised to protect Cisco confidentiality.
hoc fashion, and varied drastically across manufacturing locations. Determining the
appropriate prioritizing algorithm across all the PCBAs would ensure the PCBAs with the
yield issues and highest return on investment for yield improvements were worked on
first. This became the foundation for hypothesis 2.
Hypothesis 2: The PCBA Prioritization Algorithm
Overall PCBA yields will improve by optimally allocatingmanufacturingresources to a
holisticallyprioritizedlist ofPCBAs across the entireportfolio.
Using a Six Sigma-DMAIC approach, the PCBA PrioritizationAlgorithm was developed:
* Define the problem statement
* Measure each PCBA and assigned an index
* Analyze the list of PCBA and prioritizes each PCBA accordingly
* Improve the manufacturing yields quickly while also improving the algorithm
to ensure the most important PCBAs are worked on first
* Control and monitor PCBA yields over time
Working with a team of engineers, the algorithm to prioritize all the PCBAs was
developed. The algorithm is further discussed in Chapter 4.
2.4.
Research and Literature Review
The two hypotheses in this thesis utilize management decision science. The decision
science field is broad and has been growing across industries to enable management
teams to make optimal business decisions. Ragsdale discusses several methods in
management science such as optimization, linear programming, network modeling,
integer and nonlinear programming, and decision analysis (2004). Structured problem
solving offers more insights into the problems; and, as compared to an unstructured
approach, it is reasonable that better business decisions will occur more frequently with
higher probabilities (Ragsdale, 2004).
The next sections discuss the relevant academic literature review and industry
benchmarking for each hypotheses.
2.4.1. Component Quality Optimization Model
Based on management sciences and operations research, several engineers have
developed optimization techniques in the PCBA industry. The many papers and
models can be summarized in three broad categories:
*
Optimizing Tests
*
Optimizing Manufacturing Processes
*
Optimizing PCBA Designs
Optimizing Tests.: When optimizing tests in the PCBA manufacturing process, Stig
Oresjo analyzes the trade-offs amongst different types of structural tests (2003). The
paper discusses how PCBA complexity, manufacturing process, and test objectives
are important in determining the optimal strategy. The test strategy should account
for overall test coverage, affordability, and effectiveness. Making the proper, optimal
decisions will "result in higher quality, lower warranty, repair, and scrap costs"
(Oresjo, 2003, p 16).
Optimizing ManufacturingProcesses:Ellis, Vittes, & Kobza's discuss ways to
optimize the sequence for correctly placing components using surface mount
placement machines (2001). And, Tong, Tsung, and Yen specifically studied the
effects of using a DMAIC style approach to optimize the manufacturing solder paste
process capabilities through statistical process control and designs of experiments to
achieve a 6y performance process (2004).
Optimizing PCBA Designs: Gilbert, Bell, & Johnson propose circuit design
optimization techniques using statistical analysis and Monte Carlo simulations which
give designers visibility into how their specific design and component parameters
impact the total cost (2005). Similarly, at a different electronics company,
component defect rates were utilized during the computer aided design process to
ensure the final design would meet certain manufacturing specifications for yield and
quality.5
Although the PCBA industry utilizes several different optimization techniques to
improve test, performance, and design to improve yields, no one optimization method
focuses specifically on component selection to improve yields. Therefore, looking at
other industries, optimization techniques for selection processes can be found.
The finance industry uses optimization models when deciding which asset classes to
invest in when creating an optimal portfolio (Brealey, Myers, & Allen, 2006). With a
broad array of choices, portfolio managers need to make decisions that will allow
their portfolio to remain on the efficient frontier of increasing returns while reducing
risk. Nonlinear programming models enable the finance industry to make optimal
decisions to best maximize these returns. In the PCBA industry, choosing
components based on quality levels and costs is very similar to choosing different
assets based on returns and risk.
Moreover, a look into the automotive industry offers several best practices the PCBA
industry can utilize. "Before Toyota or Honda retains a supplier, it scrutinizes the
supplier's production process and costs structure" (Arrufiada & Vizquez, 2006, p142).
PCBA manufacturers can apply the same strategy when dealing with their suppliers.
Additionally, Toyota has worked with their suppliers to improve their processes.
Denso, for example, is responsible for much of Toyota's components, and Toyota
takes a very active role in the production process to ensure the quality level remains
high. In fact, Toyota owns a 25% share of Denso (Brooke, 2005) to ensure its
component quality levels are consistent. By working collaboratively with their
suppliers, Toyota enables its own products to meet higher quality standards demanded
by the customer. Furthermore, back in the 90s, Toyota worked with US Chip makers
to develop high quality electronic products for future automobiles (Markoff, 1990).
5Interview with design engineer from a different electronics company.
With more electronics designed into automobiles, the need to use highly reliable
components is very important. Therefore, Toyota has even helped fund a
semiconductor factory for Texas Instruments (Polluck, 1996), (Bruns, 2003) to play
an active role in managing their electronics supplier. Within the PCBA industry,
there are hundreds of suppliers. The automotive industry lends keen insights into
how to better work with suppliers to improve the overall supply chain. Perhaps, the
big PCBA players could gain more control over component quality if they actively
invested and worked with suppliers to improve design, manufacturing process, and
supply chain issues.
Fine argues that firms should utilize 3D concurrent engineering (1998) to
simultaneously design the product, the processes, and the supply chain. Doing so will
enable world class products, reinforce the companies core capabilities, and deliver
value to the customer (Fine, 1998). Additionally, industries should continue to utilize
Six Sigma and Design for Six Sigma methods earlier in the design process (Bafiuelas
& Antony, 2005). This ensures more products will meet manufacturing yield and
quality goals sooner. To improve component selection processes, optimization
methods used in the finance industry can be applied. To improve the supply chain,
PCBA firms can more actively work with their suppliers. By focusing on 3DCE the
PCBA industry can continue to better improve their product performance to ensure 60
levels are met.
In 2000, Clive Ashmore wrote an article describing how the electronics industry is
moving from SPC to dpmo as a superior means to measure defect rates. Through
various Six Sigma techniques, the PCBA industry has implemented methods and
practices to achieve better yields. In 2005, Stig Oresjo laid out a step by step test
process for determining the optimal test process. He covered how the optimal
solution really depended on "defect levels (or dpmo rates), board complexity,
manufacturing volumes, different test solutions, test effectiveness and desired quality
levels" (2005, p. 50). Just as a financial portfolio can be optimized, this thesis really
focuses in on quality levels and how selecting the right component based on quality
levels and costs can be optimized.
Moving forward, the PCBA industry can borrow practices from Toyota and move the
quality improvements up the supply chain by working collaboratively with their
suppliers. As Skinner stated in 1969, "The purpose of manufacturing is to serve the
company - to meet its needs for survival, profit, and growth" (p140). Cost, time, and
customer satisfaction (Skinner, 1969) are still very important in today's
manufacturing world. Today, 3D concurrent engineering (3DCE) summarizes these
same lessons where supply chain, product, and process improvements occur at the
same time (Fine, 1998).
The optimization will allow a firm to ensure it is on the efficient frontier for quality
and cost when selecting components while aspiring to achieve a 60 performance
process. If the firm needs help to achieve higher quality standards, the optimization
model lends itself to be a tool for firms and suppliers to work collaboratively on the
problem. Thus, the Component Quality Yield Optimization Model is a decision
making method based on 3DCE and Six Sigma principles that will positively affect
the design, process, and supply chain.
2.4.2. The PCBA Prioritization Algorithm
When seeking to understand how to prioritize hundreds of different products across
many different attributes, several types of decision making processes exist. Ragsdale
(2004) details several decision analysis techniques and methods such as:
* Multi-Attribute Utility Theory (MAUT)
* Analytical Hierarchy Process (AHP)
* Multiple Objective Decision Making (MODM)
* Multiple Attribute Decision Making (MADM)
Hundreds of research papers have been written describing the benefits of using the
decision making methods. There are many applications for using these decision
techniques. For instance, Teixeira de Almeida discusses how using the MAUT
technique in deciding which contracts to outsource (2007). Ahn & Choi use the AHP
technique to analyze a group selection of an appropriate enterprise resource planning
system where the exact solution is needed rather than a probabilistic solution (2007).
Van Hop & Tabucanon use both MODM and MODM to resolve the complexities for
the set-up problem for multiple machines in a PCB assembly line (2001). AHP and
MODM both are based on optimizing several objective functions to find the best
solutions. Shanian & Savadogo noted that a MODM analysis would offer one
solution for one function that may not be optimal for another objective function, so it
becomes difficult to find the best solution (2006).
Therefore, Shanian & Savadogo (2006) use the MADM method called the Technique
of Ranking Preferences by Similarity to the Ideal Solution (TOPSIS), introduced by
Yoon & Hwang in 1980 to select appropriate material for a fuel cell. The TOPSIS
technique works well with a finite set of attributes (Shanian & Savadogo, 2006).
TOPSIS has been used across industries as a decision making process for many
different types of problems. TOPSIS enables the proper decisions based on limited
subjectivity (Olson, 2004). According to Olson, TOPSIS provides useful decision
making techniques for several applications such as deciding which materials to use
for a specific application; where to spend manufacturing capital or make certain
financial investments; which manufacturing or robotic processes to use; and even
comparing company and financial performances (2004, p ).
The TOPSIS logic establishes a good foundation for defining the logic to rank
hundreds for types of PCBAs appropriately based on specific attributes. Each
selection is based on a set of attributes as defined by the firm. Shanian & Savadogo
offer general considerations for using TOPSIS and why I support using TOPSIS to
ranking a list of PCBAs:
*
An unlimited range of performance attributes can be included
*
Explicit trade-offs between different attributes can be accounted for
appropriately
*
The output is sorted and ranked using numerical values
* AHP Pair-wise comparisons are avoided, which is useful when dealing
with a large set of choices and attributes
* Each attribute can be weighted as defined by the firm
* The procedure is systematic, simple, and fast
In the PCBA industry, many methods are used to prioritize PCBAs with
manufacturing yield issues. Many are based on local site best practices. However, no
holistic prioritization across the firm existed, making it very difficult to know which
PCBAs in the entire product portfolio required the most resources now. Therefore,
developing a priority list for every PCBA will greatly benefit the firm. Thus, the
TOPSIS technique enables an effective approach for deciding which PCBAs are most
important to work on first.
2.5.
Summary
Chapter 2 reviewed the methodology for developing the hypotheses for this thesis. My
internship at Cisco provided the backdrop necessary to collect the data and discuss the
hypotheses in further detail. Additionally, this chapter reviewed current industry and
academic research in decision sciences pertaining to this thesis.
The following chapters will discuss the two hypotheses in further detail starting with the
Component Quality Yield Model and how different choices in components impact overall
PCBA yields. Subsequently, Chapter 4 details The PCBA PrioritizationAlgorithm.
This page has been intentionally left blank
3. Chapter 3: Hypothesis 1 Results - The Component Quality
Yield Optimization Model to achieve a 6a Yield Process
Recall Hypothesis 1: Yields and costs can be optimizedfor an entireportfolio of PCBA
products by selecting the appropriatecomponents based on component quality and cost
specifications.
When designing a product, there are vast arrays of components from which to choose in
order to meet the final product's specifications. Cisco, for example, has 196 active
product families with 23,000 product identification numbers from 600 different suppliers
and 50,000 components from around the world.6 Components are critical to the PCBA's
performance, and, therefore, the PCBA's intended use will determine which components
are used and how many components are needed. Below, Table I summarizes the most
common types of components found on any given PCBA.
Table 1: Select List of Components Used in Printed Circuit Board Assemblies
Different types PCBA Components
Application Specific Integrated Circuits (ASICs)
Board Mount Power (BMP)
Content Addressable Memory (CAM)
Capacitors
Clocks
Connectors
Data Communication Devices
Diodes
Dynamic Random Access Memory (DRAM)
Erasable Programmable Read Only Memory (EPROM)
Flash Memory
Light Emitting Diode (LED)
Linear Devices
Logic Devices
Magnetics
Microprocessors
Optical Connectors
Oscillators
Programmable Logic Devices (PLD)
Resistors
Static Random Access Memory (SRAM)
Transistors
Designers will select the proper component for the proper function based on several
criteria, such as performance and technical specifications required by the customer.
Depending on how high or low tech the product may be, any given component will vary
in both price and quality. Furthermore, once the product ramps from the design phase to
6 Cisco
documentation
high volume manufacturing, component sourcing engineers and manufacturing managers
focus on driving costs down while improving yields.
The component vendor promises a certain component quality level when supplying
Cisco. How the component performs to this quality level will affect the PCBAs that use
it. In other words, each component inherently will contribute to the overall PCBA's yield
fallout in the manufacturing process. In addition to typical manufacturing issues that
occur, component issues are culpable for a portion of yield fallout. 7 Using the "trust but
verify" mentality, a PCBA factory constantly measures yields and analyzes each failure
to determine a root cause. When a PCBA fails a test and the component is at fault, the
yield fallout is then attributed to the component. Over time the yield fallout due to that
component can be tracked and a defect per million opportunities (dpmo) rate can be
calculated. If the factory dpmo rate differs from the dpmo rate the vendor quoted, the
PCBA manufacturer will either work with the vendor to improve the quality issues or
switch vendors or components.
During a PCBA failure at a test step, additional labor costs, test equipment usage costs,
debug time costs, and component replacement costs all add to the total cost of the
PCBA's failure. This Cost of Poor Quality (COPQ) is the cost associated with a PCBA
every time it fails a test. Using components with better quality levels (or lower dpmo
rates) can improve factory yields, and thus, lower the COPQ. Likewise, lower quality
components (or higher dpmo rates) are expected to have higher component yield fallout
resulting in a higher COPQ.
To achieve a lower COPQ, the component yield fallout can be mitigated by investing
more money upfront in higher quality components. This Cost ofInvesting (COI) includes
the Cost ofImproving the Component Quality and the Cost of the Component. The Cost
ofImproving the Component Quality is overall costs associated with employees, vendors,
and suppliers working to improve the component quality. The Cost of the Component
7 The percentage of yield fallout due to component issue is not reported to protect Cisco confidentiality
looks at how overall costs will be affected by the component unit cost and usage rate
across the portfolio of products. Since higher quality components are typically more
expensive while lower quality components are cheaper, this component unit costs will
also have an affect on the overall cost of investing. Below, Table 2 summarized COPQ
and COI, the two main costs associated with components quality.
Table 2: Costs Associated with Component Quality
Different Costs
Breakdown of Cost
Costs due to the Component
Additional labor, equipment, debug,
Cost of Poor Quality Cost Spent Analyzing Additional
labor, equipment, debug,
and replacement costs all due to a
the PCBA Failure
(COPQ)
the PCBA Failure
(COPQ)
component failure
Costs working with vendors and
Cost of Improving the suppliers to improve a specific
Component Quality components quality level (e.g.
(COI)
sreduce
its dpmo rate)
Component unit cost multiplied by
Cost of the
the usage rate across the portfolio of
Component
products
Thus, the Total Component Cost (TCC), as seen in Equation 1, is the Cost of Poor
Quality plus the Cost ofInvesting. Therefore, the component selection ultimately will
affect PCBA cost and component yield fallout.
Equation 1: Total Component Cost
TCC = COPQ + COI
Consequently, design engineers, component sourcing engineers, and manufacturing
managers have an inherent tradeoff between cost and yield: They can either (1) attain an
optimal solution to achieve a minimum TCC, which will result in a certain yieldfallout,
or (2) attain an optimal solution to achieve a minimum yield fallout which will result in a
certain TCC. The component quality yield optimization model will allow design
engineers, component sourcing engineers, and manufacturing managers to run different
optimizations schemes for a portfolio of PCBAs by applying different weights to cost and
yield. These different weighting scenarios and their results are discussed further in
sections 3.3.4 and 3.4.
On one end of the spectrum, managing how design teams across multiple business units
choose the best components for new products becomes very challenging. Determining
preset component dpmo rates to achieve a 60 process will allow the PCBA manufacturers
to have a better handle on the factory yields (Ashmore, 2000). Prudent designers ought
to use a 60 component budget to determine the potential component yield fallout upfront.
On the other end of the spectrum, determining the overall impact of changing these
component quality levels that will be used across various PCBAs in current production
and the cost effectiveness of the change becomes extremely difficult.
Since similar components will have different costs and quality levels, we can assume that
different quality standards exist for a given component. Or, in other words, we can take
the current standard and elevate them several levels (Ashmore, 2000). Suppose a product
design calls for component A, which may have four different quality standards - 60,
gold, silver and bronze.8 All four standards for component A meet the technical
specifications required by the design but differ in price and quality levels as seen below
in Table 3.
8 Gold
is considered the highest quality standards available, meaning dpmo rate is the lowest. The 60
standard is considered a higher quality standard but may not be readily available to use in manufacturing.
Table 3: Component A's Quality Level 9 and Cost for Each Quality Standard
Quality
Quality Level Component Cost of Poor Cost of Investing Total Cost
Standards for (dpmo) / Overall Cost per
Quality
(COI) to switch
(TCC) =
Component A Product Yield
unit
(COPQ) from Current to COPQ + COI
Fallout
...
Current
40/5.00%
$
2.00 $
6sigma
Gold
1/ 0.08 %
10/1.20%
$
$
50.00 $
5.00 $
Silver
20/2.20%
$
Bronze
30/3.80%
$
5,000
$
-
$
5,000
250 $
2,000 $
150,000 $
25,000 $
150,250
27,000
2.50 $
4,500 $
10,000 $
14,500
1.50 $
11,000 $
5,000 $
16,000
Note: Numbers disguised to protect Cisco confidentiality
In the current state, component A has a dpmo of 40, which translates to a cost of $2.00
per unit with $5000 in COPQ for a particular PCBA. Should the management team make
a decision to switch to a different component quality standard for component A? Well,
the answer is not as simple and will depend based on how the new component standard
will affect both yield and costs for that particular PCBA and for the other PCBAs that
may also use component A.
Components B, C, D, and so on will also be used in various PCBAs. Assuming the firm
has to select a certain quality level for each component used across all the PCBAs, which
quality standards should be used for each component to achieve the highest yields and
lowest costs? Based on quality and costs variables, there is an optimal choice. But,
should this optimal decision be based on minimizing total component cost or minimizing
component yield fallout because of component quality issues?
For any given component, it may make sense to switch to a higher quality standard,
remain status quo, or even reduce to a lower quality standard. Thus, choosing amongst
the different quality standards is a much more complex decision and will have a broader
affect on overall costs and quality metrics in the entire product portfolio.
9 The 6a standard is the quality standard this particular component needs to have in order for all the PCBAs
in the portfolio of products using this particular component to meet a 6a yield process. The 6a yield
standard is further discussed in Section 3.1.
Higher quality components typically translate to higher yields and, thus, higher quality
products come with a cost premium. The question still remains: At what cost, or
tradeoff, does it make sense to improve the component quality in order to improve each
product's yield across the entire portfolio? What is the tradeoff by moving from
component A's Gold standard to the Silver standard, or to the Bronze standard? Or,
should no action be taken? How many products use this specific component? How
much will manufacturing costs improve or decline across the company? How will
overall yields be affected for the entire portfolio of products? These are questions the
Component Yield Optimization Model tried to answer.
PCBA manufactures need to be cognizant of their operation strategy, business strategy,
and core capabilities (Fine, 1998). They should be cautions about shopping for cheaper
components to lower costs and rather choose the right component for the right job to
maximize every product's yields. If companies don't adapt, they may die (Ashmore,
2000). As GM found out the hard way, choosing components solely based on component
cost may cause poorer quality in the field. Ultimately PCBA manufacturers do not want
to drive down the same path that GM notoriously paved in the 1980s.
Additionally, it is imperative to understand the implications of how component selection
will affect all product yields and costs across an entire product portfolio. Often times,
design engineers and manufacturing managers focus on their individual products rather
than taking a holistic look at how their component selections may affect other products.
Moreover, a completely different part of the business often sources the components
looking for cheaper and cheaper parts. In other words, does it make sense to invest time
and money in choosing a better quality, higher cost part to use, qualify, and implement,
or will the current approved component suffice? Should the component sourcing team
continue to nickel and dime the component vendors in search for cheaper components?
Furthermore, due to the nature of contracts where many include a minimum order
quantity or reduce cost when bought in bulk, the decision to source a new component will
most likely affect more than one product's yields.
Imagine Component A is used in only 5 PCBA's but Component B is used in 500
PCBAs. Deciding which quality standards to use for which components across multiple
PCBAs becomes an extraordinarily challenging task. A nonlinear optimization program
will aid in making these challenging decisions. The model will be based on the COPQ
and the COI concepts discussed earlier.
In the subsequent sections, this chapter fully details the component quality yield
optimization model. This is a proof of concept version and will explain how the model
will theoretically work enabling design engineers, component sourcing engineers, and
manufacturing managers to make better component selection decisions in order to
minimize costs or component yield fallout. 10 The method is based on three main steps
summarized below:
1. Define the 60 component quality standard: Define what the component dpmo
rates need to be to achieve a 6cr yield process for every PCBA in the entire
portfolio.
2. Calculate the Total Cost of the Component: The Cost ofPoor Quality can be
derived using factory yield data and validated component yield fallout. The Cost
ofInvesting can be calculated by determining the Cost of mproving Component
Quality, which is based on the incremental costs needed to achieve the new dpmo
rates and the known Cost of the Component for each quality standard.
3. Optimize the component selection across the entire PCBA portfolio by selecting
the proper component quality standards for each component based on minimizing
the Total Component Cost or minimizing the Component Yield Fallout.
The optimization model does not consider the cost of field failures such as immediate
returns, dead-on-arrivals, or regular customer returns. Field failure data is available and
very important, but correlating the cost of failures to the component level is out of the
scope of this thesis.
10 To protect Cisco, numbers used in this chapter are disguised or entirely made up. What's more important
is the methodology for developing and using the model.
Because once the component quality level is decided and sourced, this component will be
used across multiple PCBAs. The component quality selection will have a broad impact
on yield and cost. Assuming designer engineers, component sourcing engineers, and
manufacturing managers have a range of quality levels from which to choose for each
component, and each quality level meets the required performance and technical
specifications determined by the customer, then optimizing yield and cost metrics will
drive the final decision for which component quality levels to use. The component
quality yield optimization model is a tool that will allow managers to make better
business decisions to optimize yields and costs in the factory by better understanding the
cost impact and yield improvement of changing from one component quality level to
another.
As noted earlier, there is an inherent tradeoff between Total Component Cost and
Component Yield Fallout. In the end it is the firm's decision to choose which tradeoff to
pursue based on its current operations strategy. However, by employing the component
quality yield optimization model, the firm will be one step closer to achieving a 6G yield
process.
Before identifying which component quality levels to use, sections 3.1 to 3.3 step through
how the component quality yield optimization model is built.
3.1.
Introduction to PCBA Yield Calculation
Manufacturing companies use yields to measure the performance and quality of the
manufacturing process. Achieving and maintaining high yields will remain important in
any manufacturing environment. Improving printed circuit board assembly yields is
becoming increasingly critical to remain competitive. PCBA yields can be calculated in
many different ways. As discussed previously in Section 1.2, yields are typically
separated into two main categories: structuralandfunctional. The quality of an
individual component will affect the functionality of the PCBA, not the structural build.
Therefore, the component quality yield optimization model will only look at the effect on
functional yields as shown below in Equation 2.
Equation 2: Functional Yield" Equation for any PCBA
FY = 100% -C-0,
where
FY = functionalyield %
C = component yieldfallout %
0 = other yieldfallout %
There are many other factors that cause yield fallout; however, this thesis investigates
yield fallout due to component issues only. Equation 2 assumes that all other factors
affecting yield fallout are achieving a 60 yield process, and if the component yield fallout
variable achieves a 60 process, then overall functional yields will achieve a 6c yield
process. Therefore, in order to achieve a 60 functional yield process for any given
PCBA, it is necessary to know where each component is used and what each component's
yield fallout must be in order to reach the 60 goal.
In short, the component quality yield optimization model will perform two main
functions. First, based on the component usage rate across the entire portfolio, the
optimization model will determine what each component dpmo rate needs to be to
achieve a 60 yield process for every PCBA. Second, the model will determine which
components quality levels (60, Gold, Silver, or Bronze) should be selected, based on
yield and cost improvements. Before delving into the model, however, the following
section discusses how the optimization model predicts the component yield fallout for
each PCBA based on the component's quality level.
3.2.
ProbabilisticYield Model based on Components Defect Rates
Every PCBA manufactured has a different mix of component type and count. Depending
on the complexity of the board, there may be fewer than 100 components or more than
11The entire yield fallout equation is disguised to protect Cisco confidentiality.
5000 components. By analyzing the bill of materials for each PCBA, the specific
component usage rate can be determined. Additionally, each component has an
associated quality level measured in defects per million opportunities, or dpmo rates.
Therefore, the overall PCBA yield fallout caused by the components can be determined
by using the bill of materials combined with the dpmo rates for every component.
From Equation 2, it is necessary to determine the component yield fallout on any given
PCBA. Using a probabilistic model with corresponding component dpmo rates, the
predicted component yield fallout percentage, or C, for any given PCBA can be
determined. First, the probability that the PCBA will not fail 12 the functional test is
calculated. Assuming that each component is independent and any one component
failure will cause the entire PCBA to fail, the probability of the PCBA failing is the
probability that any given component on the PCBA will fail. Since the dpmo rate is
known for each component, this dpmo rate can be used to represent the probability of
PCBA failure because of that specific component. Therefore, for a given component i,
the probability that the PCBA does not fail due to the ith component can be rewritten as
follows in Equation 3.
Equation 3: Probability of PCBA Not Failing due to the ih Component
P(PCBA does not Fail) = [1 - P(PCBA Fails due to the it' component)] m
= [1 - i th component's dpmo]
m
where m = the number of i h components used in the PCBA
Therefore, taking the product of all the probabilities of not failing across all the different
components used will determine the probability that the PCBA will be successful.
Finally, to calculate C, the resulting product of all the probabilities that determine the
PCBA's success is subtracted from one and multiplied by 100. Determining C using the
probabilistic yield model is shown in Equation 4.
12 The
probability of not failing a test due to a component issue can also be written as the probability of
successfully passing the test and yielding 100%.
Equation 4: Calculating the Component Yield Fallout% using a Probabilistic Model
V any given PCBA,
C = [1- [1- P(PCBA Fails due to components) ]]x 100
= I -I [1 - i'h component dpmol]
x 100
where m = the number ofi th components used in the PCBA
n = the total number of components used on the PCBA
Equation 4 represents the probabilistic yield model, and it is the foundation for the
component quality yield optimization model and also the reason why the program is
nonlinear.
3.3.
The Component Quality Yield Optimization Model
The component quality yield optimization model is based on the probabilistic yield
fallout equation described in section 3.2 and works in three main steps. First, excluding
any costs, using a nonlinear continuous optimization program the model determines the
optimal 6a dpmo rates, or 60 dpmo budgets, for each component so that all PCBAs in the
portfolio can achieve a 6o yield level. Second, a component quality expert calculates
incremental costs to achieve the component dpmo budget. As discussed earlier, the Total
Component Costs are incurred in two different ways as seen in Equation 5.
Equation 5: PCBA TCC Equation
TCC = _1 COPQ + Z2 COI
First, the COPQ is the cost associated with yield fallout in the factory caused by
component quality. COPQoccurs when a PCBA fails a specific test and then requires
certain resources to identify and fix the problem. Since each component has a dpmo rate,
each component has a COPQcaused by the component yield fallout.
The second cost is associated with investing money to achieve the proposed dpmo
budget, or the Cost oflInvestment (COI). As described earlier, the COI can also be
broken down into the Cost of Improving the Component Quality and the Cost of the
Component. This cost is incremental which covers the human capital and other resources
needed to work with the vendors to improve the quality levels of the current component
or to source an entirely different component with a better quality level from a different
supplier. COI also includes the cost associated with using the component selected and is
detailed in Equation 6.
Equation 6: Cost of Investment
12 COl = 12a Cost of Improving the Component Quality + 12b Cost of the Component
Since one component can be used across many different PCBAs, it is necessary to
understand what component dpmo rates need to be to enable 60 yields for every PCBA
manufactured. Because the 60 numbers can be quite aggressive, three other quality
standards are introduced: a gold, silver, and bronze standard. Gold standard is considered
the highest quality standard available while the bronze standard is the lowest quality
standard available. The optimization model will also account for the current, "As-Is"
standard using the dpmo rates in the current manufacturing process.
To use the model, it is necessary to quantify the effect of reducing component dpmo on
functional yields, determine the costs associated with the reduction, and then optimize
which component quality levels should be selected to either minimize cost or minimize
yield fallout.
The component quality yield model will follow these steps:
1. Use a nonlinear continuous optimization program to define new component 6C
dpmo budgets to achieve a 6( yield for every PCBA
2. Calculate incremental TCC as follows:
*
Calculate COPQ caused by component yield fallout
*
Calculate COI to achieve proposed dpmo budgets
3. Use a nonlinear binary discrete optimization program to select which component
quality levels should be used by minimizing TCC or minimizing C.
The model assumes that all quality levels for all the components can be used on any of
the PCBAs manufactured and each PCBA's yield fallout is determined by the
component's dpmo and usage rates. The component quality yield optimization model
investigates how different component selections will affect the Component Yield Fallout
Percentage(C) and Total Component Cost (TCC) for every PCBA and for the overall
portfolio of products. Thus, the tradeoff surfaces again, and the firm can either (1)
optimize the TCC to achieve a minimum which will result in a certain C, or (2) optimize
the C to achieve a minimum which will result in a certain TCC. The following section
discusses each step of the component quality yield optimization model.
3.3.1. Determine the 6a Budgets for Each Component
From Equation 2, each PCBA's 6a yield goal is defined by multiple factors including
component yield fallout. Therefore, knowing what the 6a yield equation can absorb
for component yield fallout will determine the 6a budgets for each component. The
nonlinear continuous optimization program is set up as a penalty function and will
minimize the error (A)between the component yield fallout percentage (C), and the
60 yield fallout % goal (G). Therefore the objective function will minimize A2 to
allow for solutions where the optimal solution is better than G.
Objective Function: Minimize A2
where A =
G PCBA,i -
CCBA,i
) and where
G = 6r component yieldfallout % goal
C = Component Yield Fallout %
By changing:
* Component dpmo budget values for each component (see Table 4)
Table 4: Decision Variables to Determine 60 Component dpmo Budget
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
"As-Is" Current
dpmo
78
27
23
8
3
48
7
6
22
14
10
58
85
79
49
75
4
11
67
98
91
72
49
Note: Numbers disguised to protect Cisco confidentiality
Subject to:
* No non-negative values for dpmo rates are allowed
*
CPCBA,i 5 GPCBA,i (see Table 5)
Table 5: Constraints for Determining 6a Component Budgets by Minimizing A2
PCBA exam Dies
PCBA#1
PCBA#2
PCBA#3
PCBA#4
PCBA#5
Sum of Yield Fallout
C, Predicted
Component Yield
Fallout with dpmo
budaets
I
G, 6a Component
Yield Fallout %
Goal
5
0.13%
A
A2
0.13%
0.00%
<
5
0.19%
0.30%
0.19%
0.30%
0.00%
0.00%
5
0.24%
0.24%
0.00%
<
0.23%
1.09%
0.23%
1.09%
0.00%
0.01%
Note: Numbers disguised to protect Cisco confidentiality
This optimization is nonlinear because of the nature of using a probabilistic model to
determine the yield fallout as defined earlier in Equation 4. The optimization model
calculates what each component dpmo needs to be in order to achieve a 6a yield for
every PCBA in the portfolio. After running the 60 budget optimization across an
array of five randomly selected PCBAs, each component's 60 dpmo budget is
determined as seen in Table 6.13
13 PCBAs are disguised to protect Cisco confidentiality
Table 6: 6a dpmo Budgets so Every PCBA Achieves a 60 Yield
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
"As-Is" Current
dpmo
78
27
23
8
3
48
7
6
22
14
10
58
85
79
49
75
4
11
67
98
91
72
49
60 Standard dpmo
1
1
1
1
1
0
1
1
1
1
3
1
0
0
1
0
0
0
298
33
0
1
0
Note: Numbers disguised to protect Cisco confidentiality
Right away the program determines that the 60 budgets to be very low to meet the 6G
yield goals. In fact, most 60 budgets are driven to zero while the nonlinear nature of the
program homes in on certain components in the portfolio and adjusts it to allow the
program to meet all the constraints. 14 By using these 6c dpmo budgets, all five PCBAs in
the portfolio can then meet their 60 yield goals within a few small percentage points, and
the entire portfolio meets its goal as well.
Table 7 and Table 8 show a comparison of how the "As-Is" and 6G dpmo budgets affect
overall yield fallout due to the differences in component quality levels. Note the higher
component yield fallout percentages in PCBA#3 and PCBA#4 are driven by the
14 Please
note that this particular solution may not be unique. Since the optimization program is
nonlinear,
several optimization runs will need to be tested to determine the optimal solution.
complexity of the board and the type of components used. These two particular PCBAs
have an extremely higher component count than PCBA#1, #2, and #5.
Table 7: Summary of "As-Is" dpmo Goals for 5 PCBA's Component Yield Fallout
"As-Is"
PCBA examples
PCBA#1
PCBA#2
PCBA#3
PCBA#4
PCBA#5
Sum of Yield Fallout
C, Predicted
Component Yield
Fallout with "As-Is"
dpmo budgets
6.88%
7.48%
32.07%
29.64%
9.28%
85.35%
5
5
5
5
<
G, 60 Component
Yield Fallout %
Goal
0.13%
0.19%
0.30%
0.24%
0.23%
1.09%
A
-6.75%
-7.29%
-31.77%
-29.40%
-9.05%
-84.26%
A2
0.46%
0.53%
10.09%
8.64%
0.82%
71.00%
Note: Numbers disguised to protect Cisco confidentiality
Table 8: Summary of the 6a dpmo Budgets for 5 PCBA's Component Yield Fallout
C, Predicted
Component Yield
6a
Fallout with 60 dpmo
PCBA examples
budgets
PCBA#1
0.12%
PCBA#2
0.17%
PCBA#3
0.30%
PCBA#4
0.23%
PCBA#5
0.24%
Sum of Yield Fallout
1.06%
5
5
s
5
5
G, 6a Component
Yield Fallout %
Goal
0.13%
0.19%
0.30%
0.24%
0.23%
1.09%
A
0.01%
0.02%
0.00%
0.02%
-0.01%
0.03%
A2
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
Note: Numbers disguised to protect Cisco confidentiality
In this theoretical example, there is a substantial difference between the "As-Is" dpmo
and the 6a dpmo budget results. Since the 6a dpmo budgets are very close to zero,
attaining these dpmo levels may be next to impossible or extremely expensive. This
optimization to determine the 6a budgets is run without a TCC variable. Therefore, this
thesis will analyze additional options adjusted for different quality and cost levels. To do
so, three different quality levels for each component will be created. As seen in Table 9,
the gold standard will be the highest quality available, followed by the silver standard and
then the bronze standard. The 6a standard and the "As-Is" standard will also be
considered when determining which components to use. The "As-Is" dpmo standard
represents the current quality level for each component at Cisco. The 60 Standard dpmo
is determined from the nonlinear continuous optimization program in section 3.3.1.
Table 9: Different Quality (dpmo) Levels for Each Component
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
"As-Is" Current 6a Standard
dpmo
dpmo
78
1
1
27
23
1
8
1
1
3
0
48
7
1
1
6
22
1
1
14
3
10
1
58
0
85
0
79
1
49
0
75
0
4
0
11
298
67
33
98
91
0
1
72
49
0
Gold Standard
Silver Standard
Component
dpmo
Component dpmo
19
39
14
7
11
6
2
4
2
1
24
12
2
3
3
1
11
5
7
4
5
3
29
14
43
21
40
20
24
12
19
38
2
1
6
3
17
34
49
24
23
45
36
18
25
12
Bronze
Standard
Component
dpmo
58
20
17
6
2
36
5
4
16
11
8
43
64
60
37
56
3
8
51
73
68
54
37
Note: Numbers disguised to protect Cisco confidentiality
The remaining Gold, Silver, and Bronze quality standards are derived in three main ways.
The first method uses known component dpmo rates for the components that already
have a variety of quality levels from which to choose. Certain vendors already carry
different quality levels for certain components, or, certain quality levels of components
can be sourced from various vendors. The second method is based on incremental
improvements the firm sets for each component dpmo rate. In these cases, the 60
Standard dpmo is the stretch goal, while the Bronze, Silver, and Gold Standards represent
incremental quality improvements the firm could realize quarter over quarter or year over
year. The last method to determine the different quality levels for each component is by
forecasting potential dpmo rates where they do not exist. Working with different
component vendors and component quality experts, it is possible to estimate what dpmo
rates could be for each quality level.
Based on these five different quality levels in Table 9, the following sections continue
with the component quality yield optimization model and set up the second nonlinear
optimization program to determine when it is best to switch to a higher quality, lower
dpmo component. To properly optimize the results across the entire portfolio of PCBAs,
costs need to be fleshed out in more detail.
3.3.2. Calculated the Cost of Poor Quality (E1) for each PCBA
Every time a PCBA fails because of a component issue, there is a cost associated with
this failure. Recall that each PCBA has an associated COPQ and is represented by E1
from Equation 5. This model uses the COPQ caused by the component yield fallout as
predicted by the probabilistic model. By using the bill of materials for every PCBA, a
COPQ for component yield fallout can be calculated. When different components are
used, each PCBA will incur a different yield fallout, and, thus, realize a different COPQ.
By using high quality, low dpmo components, the COPQ will be low. Likewise, using
low quality, high dpmo components will drive the COPQ higher. Therefore, the COPQ is
an important cost driver when determining which quality components to use.
3.3.3. Calculated the Cost of Investment (E2)
Another important cost driver is determining how much it will cost to either switch to a
new component or invest money to drive the current component's dpmo to a higher
quality standard. The COI (C2) from equation 6 includes the Cost of mproving the
Component Quality (Z2) and the Cost of the Component (E2b). Additionally, investment
costs include but are not limited to the time and money the firm uses to improve the
component dpmo such as employee labor hours, working with the vendors, travel time,
etc. In the same manner that the different quality levels for each component is
determined (see section 3.3.1), the costs for each quality level are derived using similar
methods: (1) Use available data where possible, (2) Base the costs on company stretch
and incremental goals, and (3) Estimate costs by working with vendors and component
quality experts. The COI (E2) for each component quality level is summarized for E2a in
Table 10 and E2b in Table 11.'5
Table 10: Cost of Investment,
2,Cost
of Improving the Component Quality
Cot of Investment (COI a) : E2.
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
Investment Cost to
Stay "As-Is" using
Current dpmo rates
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
-
Investment Cost ot
Achieve 6o
Standard DPMO
13,370
$
$
16,697
3,161
$
3,520
$
17,574
$
3,777
$
19,897
$
9,042
$
$
2,845
$
7,008
$
19,421
14,370
$
10,220
$
7,835
$
2,904
$
$
18,527
1,736
$
23,574
$
8,380
$
3,770
$
$
10,013
23,084
$
$
16,127
Investment Cost
to Achieve Gold
Standard
$
4,178
$
501
988
$
$
456
$
5,492
$
1,180
$
6,218
$
2,826
$
889
$
663
6,069
$
$
4,490
3,194
$
$
493
2,601
$
3,400
$
$
542
$
7,367
$
2,619
$
1,178
3,129
$
$
5,200
5,040
$
Investment
Cost to
Achieve Silver
Standard
$
2,786
$
384
$
450
533
$
730
$
787
$
$
1,500
$
126
44
$
$
34
$
4,046
2,994
$
2,129
$
356
$
1,734
$
$
1,200
$
362
$
4,911
1,746
$
$
489
$
1,289
4,809
$
$
3,360
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
Investment
Cost to
Achieve
Bronze
Standard
1,393
132
329
367
654
160
1,022
56
20
3
2,023
1,497
1,065
61
867
89
181
120
873
393
1,043
2,405
800
Note: Numbers disguised to protect Cisco confidentiality
On one end of the spectrum, staying with the current components' dpmo rates, the "As-Is"
method costs no money and represents the lowest COI. On the other end, the highest
COI is to switch to the 6y standard. The investment cost to switch to gold, silver, and
bronze standards fall somewhere in between the two extremes.
15 All
cost figures and numbers are disguised to protect Cisco confidentiality.
Table 11: Cost of Investment,
22b
Cost of the Component
Cost of the Investment (COI b) :E•b
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
Unit Cost to Stay
"As-Is"
4.6
$
4.3
$
0.8
$
3.5
$
2.7
$
4.4
$
2.9
$
2.7
$
3.3
$
1.1
$
1.3
$
0.4
$
0.9
$
1.8
$
0.4
$
0.2
$
3.6
$
1.2
$
1.5
$
3.3
$
2.8
$
3.7
$
4.3
$
Unit Cost for
the 6a
Standard
17.7
$
$
5.2
15.0
$
$
12.2
13.7
$
19.4
$
7.0
$
$
13.1
11.2
$
$
9.8
7.6
$
17.9
$
$
13.4
9.8
$
$
3.9
10.2
$
16.2
$
$
4.9
11.9
$
$
14.5
8.3
$
1.8
$
10.1
$
Unit Cost for
the Gold
Standard
6.03
$
3.35
$
10.96
$
$
4.45
9.73
$
12.35
$
5.12
$
$
9.71
7.58
$
7.99
$
5.16
$
12.01
$
$
8.98
3.66
$
3.24
$
2.66
$
11.70
$
$
4.84
7.78
$
9.01
$
6.53
$
1.10
$
6.62
$
Unit Cost for
the Silver
Standard
4.10
$
3.06
$
$
3.26
$
3.40
3.03
$
4.12
$
1.56
$
$
1.20
0.76
$
2.20
$
$
4.51
$
10.56
$
7.92
2.37
$
2.30
$
1.00
$
9.53
$
$
2.87
7.03
$
4.23
$
$
1.78
1.04
$
$
5.94
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
$
Unit Cost foi
the Bronze
Standard
3.54
1.04
3.01
2.45
2.74
2.30
1.39
0.98
0.26
0.16
1.53
1.35
1.66
1.97
0.78
0.22
1.89
0.54
2.39
2.91
1.66
0.35
1.74
Note: Numbers disguised to protect Cisco confidentiality
Now that the COPQ and COI are known, the next step is to optimize which
components to use.
3.3.4. Set up the Nonlinear Program and Optimize
Based on the five component quality levels available and knowing the total cost, a
nonlinear binary discrete optimization model can determine which quality levels of
components to use by either (1) minimizing TCC 16 which will result in a certain C' 7,
or (2) minimizing C will result in a certain TCC.
Each component is segmented into predetermined quality levels as seen in Table 9.
The optimization program will choose only one type of quality level to use for each
'6 TCC = Total Component Cost = (COPQ + COI)
17 C = Component Yield Fallout%
component based on the constraints and the objective function. The objective
function is set up to allow different weights on either the TCC or C as follows:
w1 Z TCC + w2 I C. Because the 6c standard component dpmo is an option to
choose as a quality level, this objective function inherently incorporates the penalty
function from the first optimization program discussed in section 3.3.1. Five main
weighting scenarios will be optimized and compared against five different preset
component quality selections as defined below in Table 12.
Table 12: Weighting Scenarios for the Objective Function
Scenario
I
II
III
Description
Focus on TCC only
Focus on C only
Equal Weights on TCC and C
W1
1.0
1.0
0.5
w2
0.0
0.0
0.5
IV
V
Weight TCC
Weight C
0.9
0.1
0.1
0.9
Scenario Preset Component Quality Selections
VI
Use all 60 Components
VII
Use all Gold Components
VIII
IX
Use all Silver Components
Use all Bronze Components
X
Use all "As-Is" Components
Scenario I- V uses the nonlinear binary discrete optimization program. Each
optimization run for ScenariosI- V is set up as follows:
Objective Function: Minimize w, I TCC + w2 I C where wl and w2 will be
determined by Table 12.
By changing:
*
Choose either 60, Gold, Silver, Bronze, or "As-Is" Standard
represented by 60, G, S, B, and AI, respectively. (see Table 13)
Subject to:
*
Non-negative values are not allowed for dpmo rates
*
The use of 6y, G, S, B, and AI are binary decision variables where 1
indicates the level is used and 0 indicates the level is not used.
60 i + Gi + Si + Bi + Ali = 1, for each ithcomponent, that is, only one
value for each component can be selected.
Table 13: Decision Variables for Optimization Model
Use Validated
DPMO
(As-Is)
Decision Variables
Use Gold
Use 6a
Use Silver
Use Bronze
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
Note: Numbers disguised to protect Cisco confidentiality
The model only allows the choice of one type of quality level for each component.
By choosing a specific quality level, the resulting TCC and C are calculated for each
PCBA and the entire portfolio.
Scenarios VI - X were analyzed by manually selecting the specific quality levels for
each component and interpreting the resulting TCC and C values.
The next section will detail the results from the optimization runs for these ten
scenarios.
3.4.
Results and Application of the Component Quality Yield
Optimization Model
This section will look at the results of running the component quality yield optimization
model on ScenariosI- V as well as the results from the manually produced Scenarios VI
- X (see Table 12). The model is run using Microsoft Excel solver based on the five
component quality levels previously discussed in Table 9.
In Scenario I, the optimization model only minimizes the TCC since wi equals 1 and w 2
equals 0. Minimizing TCC results in a certain C. The component selection results can be
seen in Table 14, where a "1" represents the selection of a certain quality standard for a
specific component.
Table 14: Scenario I's Component Selection Results
Decision Variables
Use Validated
DPMO
(As-Is)
Components
Component A
Component B
Component C
Component D
Coomponen
Use 6a
Use Gold
1
I
Component F
Component
Component
Component
Comoonent
Component
(G
H
I
J
L
Use Silver
Use Bronze
1
--
Component M
C-
omIII
Ionent
I•N
Component O
Component P
Co.nmnennt Q
Component R
Component S
Component T
Component U
Com onent V
Component W
Component X
Note: Numbers disguised to protect Cisco confidentiality
I
In this proof of concept example, the optimization program selected a range of different
component quality levels depending on the component. For this example, the lowest
achievable TCC that meets all the constraints is $187,351 with a C of 38.5% across the
entire portfolio of five PCBAs. The data summary for Scenario I is shown in Table 15,
and the component yield fallout percentage by PCBA is below in Table 16.
Table 15: Summary of Scenario I's Optimization Results
Parameters
-1Cost of Poor Quality (COPQ)
72 Cost of Invesment (COI)
12a Cost of Improving the Component Quality
-2b Cost of the Component
+
Total Cost = 7, 72
7 of C for all PCBAs across Portfolio
7 of G for all PCBAs across Portfolio
A
A2
Potential PCBA Yield =
(1 - Eof C for all PCBAs across Portfolio)
$
$
$
$
$
Values
129,264
58,087
13,001
45,086
187,351
38.47%
1.22%
37.25%
13.88%
61.53%
Table 16: Scenario I's Component Yield Fallout Percentage by PCBA
Scenario I
PCBA#1
PCBA#2
PCBA#3
PCBA#4
PCBA#5
Sum of Yield Fallout
C, Predicted
G, 6o
Component Yield Component Yield
Fallout % Goal
Fallout %
3.17%
0.25%
3.50%
0.25%
14.11%
0.24%
12.81%
0.24%
4.88%
0.25%
38.47%
1.22%
A
2.93%
3.26%
13.87%
12.57%
4.63%
37.25%
A2
0.09%
0.11%
1.92%
1.58%
0.21%
13.88%
Note: Numbers disguised to protect Cisco confidentiality
Although Scenario I minimizes the TCC, the C is still quite high and is off by 37.25%
from the 6a yield fallout percentage goals. The 6o yield fallout percentage goal, or G, is
determined by the proprietary yield formula which sets the yield goal for each PCBA.
Recall from Table 8 in section 3.3.1, using all 6a quality level components gives a 1.06%
yield fallout. How much would it cost to achieve the 60 levels? Is it feasible to even
attain this quality level?
In Scenario II, the optimization model will minimize C by setting wl equal to 0 and w2
equal to 1, which results in a certain TCC. Table 17 shows which quality levels need to
be selected for each component to minimize C.
Table 17: Scenario II's Component Selection Results
Use Validated
DPMO
Decision Variables
Use Gold
60
Use
Use 60a
Use Silver
Use Bronze
(As-Is)
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
I
O
Component
Component
Component
Component
Component
Component
Component
Component
Component
Component
P
Q
R
S
T
U
V
W
X
'I
~-----
I
I
I
I
I
I!
Note: Numbers disguised to protect Cisco confidentiality
Different combinations of component quality levels push C to a level of 3.10% for the
entire portfolio. Scenario II's data summary is shown in Table 18.
Table 18: Summary of Scenario II's Optimization Results
Parameters
E1 Cost of Poor Quality (COPQ)
72 Cost of Invesment (COI)
12a Cost of Improving the Component Quality
72b Cost of the Component
Total Cost = 1,+ -2
7 of C for all PCBAs across Portfolio
7 of G for all PCBAs across Portfolio
A
A2
Potential PCBA Yield =
(1 - 7 of C for all PCBAs across Portfolio)
$
$
$
$
$
Values
18,948
245,302
57,351
187,951
264,250
3.10%
1.22%
1.88%
0.04%
96.90%
This yield fallout value differs by 1.88% when compared to the 60 yield fallout
percentage goals but comes with a total component cost of $264,250. Scenario II
represents a 35.4% yield improvement for an additional $76,900 over Scenario I.
Additionally, each PCBA's C comes closer to the 6a level as seen in Table 19.
Table 19: Scenario II's Component Yield Fallout Percentage by PCBA
Scenario II
PCBA#1
PCBA#2
PCBA#3
PCBA#4
PCBA#5
Sum of Yield Fallout
C, Predicted
Component Yield
Fallout %
0.25%
0.39%
1.20%
0.82%
0.43%
3.10%
G, 6a
Component Yield
Fallout %Goal
0.25%
0.25%
0.24%
0.24%
0.25%
1.22%
A
A2
0.01%
0.15%
0.96%
0.58%
0.19%
1.88%
0.00%
0.00%
0.01%
0.00%
0.00%
0.04%
Note: Numbers disguised to protect Cisco confidentiality
The optimization model is run across Scenarios III - V in the same manner. The results
for the five different weightings are shown in Table 20.
Table 20: Summary of the Optimization Results for Scenarios I - V
Summary
Scenario II
I, Cost of Poor Quality (COPQ)
Z2
$
$
Cost of Invesment (COI)
Cost of Improving the $
Component Quality
Z2b Cost of the Component $
Z2a
TCC = Z1 + Z2
Z of C for all PCBAs across
$
Portfolio
7 of G for all PCBAs across
Portfolio
A
A2
Potential PCBA Yield =
(1 - 7 of C for all PCBAs across
Portfolio)
$
58,087 $
129,264
Scenario V
Scenario Ill
Table
for
I
Sce
Scenario II
Scenario III
18,948 $
104,616
245,302 $
74,581
-
V
Scenario IV
152,577
$
36,504
$
$
$
Scenario V
29,386
218,300
$
57,351 $
11,030
$
9,409
$
43,662
45,086 $
187,351 $
187,951 $
264,250 $
63,550
179,196
$
$
27,095
189,081
$
$
174,638
247,686
13,001
38.47%
3.10%
27.88%
53.70%
4.79%
1.22%
1.22%
1.22%
1.22%
1.22%
37.25%
13.88%
1.88%
0.04%
26.65%
7.10%
52.48%
27.54%
3.57%
0.13%
61.53%
96.90%
72.12%
46.30%
14.65%
Note: Numbers disguised to protect Cisco confidentiality
Additionally, Scenarios IV - X were analyzed. These scenarios are predetermined by
manually selecting the component quality level. For instance, to analyze Scenario VI,
"l s" were typed into the "Use 6G" column as seen in Table 21.
Table 21: Example of Manually Selecting All 6g Components
Use Validated
DPMO
(As-Is)
Components
Component A
Component B
Component C
Component D
Component E
Component F
Component G
Component H
Component I
Component J
Component L
Component M
Component N
Component O
Component P
Component Q
Component R
Component S
Component T
Component U
Component V
Component W
Component X
Decision Variables
Use 6a
Use Gold
I
Note: Numbers disguised to protect Cisco confidentiality
Use Silver I Use Bronze
Using all 6a dpmo quality levels for every component generates a C of 1.06% across the
entire portfolio at a TCC of $658,712. It is worth noting that because of Excel solver's
limitations, the nonlinear optimization program did not reach this specific minimum
1.06% component yield fallout earlier in Scenario II. When selecting all 60 components,
the lowest C is achieved. Scenario II should determine this solution but rather
determined a solution very close. For example, Scenario II's C is 3.10%, only 2.04%
worse than Scenario VI's C.18
Similarly, using all Bronze quality level components gives a yield fallout of 66.27% for
$209,139. Table 22 summarize the results of all the five predetermined scenarios in a
side by side comparison. To illustrate the main differences in all ten scenarios, the results
are also displayed graphically in Figure 3.
Table 22: Summary of the Results for Scenarios VI - X
Summary Table for Scenarios VI - X
,1 Cost of Poor Quality (COPQ)
-2 Cost of Invesment (COI)
E2a Cost of Improving the
Scenario VI
7,514
$
$
651,197
$
Component Quality
E2b Cost of the Componeni $
TCC = E, + -2
$
Z of C for all PCBAs across
Portfolio
7 of G for all PCBAs across
Portfolio
A
Potential PCBA Yield =
(1 - Eof C for all PCBAs across
Portfolio)
Scenario VII
$
$
Scenario VIII
138,423
91,937 $
195,086
$
111,769
$
Scenario IX
167,074
$
$
Scenario X
186,668
42,064
$
23,086
256,851
$
68,712
$
36,798
$
15,551
$
394,346
658,712
$
$
126,373
287,023
$
$
74,972
250,192
45.77%
$
$
26,514
209,139
$
$
1.06%
23.73%
66.27%
23,086
209,755
85.35%
1.22%
1.22%
1.22%
22.50%
5.06%
1.22%
44.55%
19.84%
1.22%
-0.16%
0.00%
65.04%
42.31%
84.13%
70.78%
98.94%
76.27%
54.23%
33.73%
14.65%
Note: Numbers disguised to protect Cisco confidentiality
18Improving
component yield fallout percentage by 2.04% from Scenario VI to Scenario II, would require
an additional total component cost of $394,462.
Figure 3: Total Cost Curve and Associated Yield Fallout for Different Component Selections
90.00%
$700,000
80.00% o (1
0
$600,000
*
70.00%
S$500,000
o1
60.00% =1
>o
50.00%
)
$400,000
IW
$300,000
o1.
o
40.00%
I-
0
C.
3L
-0
Z
0M -
30.00%
0
$200,000
10.00%
$100,000
S0.00%
5:
>
=-
13
=o
4o)
.o
I)O
U)
V~
(I)
VI
(II
rn
--
Total Component Cost (TCC)
--
Projected Component Yield Fallout%, C Different Component DPMO combinations
Using the data from Table 20 and Table 22, an efficient frontier can be drawn by
analyzing Potential PCBA Yields for a certain cost. Potential PCBA Yields are
calculated as 100% - C and the efficient frontier is shown in Figure 4. The goal ought to
be to achieve the highest Potential PCBA Yield for the lowest TCC.
Figure 4: PCBA Yield - Cost Efficient Frontier
Scenario Vil
100.00%-
-- ----- " '
----------cr-'-~iT
I I
1ImZdlttJ~lil^
......
90.00%80.00%-
V
r
/1
Scenario III
D Scenario vI I
I
70.00%-
.
60.00%450.00%-
Q
'•
30.00%
S
dl
se Scenario I
A
I
Scenario VIII
SScenario IV
Scenario IX
20.00%Scenario X
Scenario V
10.00%0.00%
$-
-
- -
$100,000
Efficient Frontier
$200,000
$300,000
$400,000
$500,000
$600,000
Total Component Cost (TCC) for Entire Portfolio
$700,000
Figure 4 illustrates the inherent tradeoff that occurs when ultimately deciding which
quality level of components to select. Each combination of component quality levels will
affect the TCC and C for each PCBA and ultimately the entire portfolio. This component
quality yield model suggests that the current "As-Is" dpmo levels (Scenario X) is close to
the efficient frontier but not the most optimal solution. By switching from the "As-Is"
(Scenario X) to Scenario I, II, III, IV, or even IX, the firm will be maximizing its
Potential PCBA Yield for a similar or even better TCC.
The all Gold, Silver, and Bronze component choices (Scenario VII, VIII, and IX) each
incur higher costs for better yields, but these scenarios still do not lie on the efficient
frontier. Switching to these combinations would not be the most optimal. The all 6a
component quality decision (Scenario VI) does lie on the efficient frontier but comes at a
very high TCC. Scenario II and III lie closest on the efficient frontier and return the
biggest potential yield improvements for the least amount of investment. These two
scenarios represent the combinations of different component quality levels that
simultaneously drive Potential PCBA Yields up for the best TCC, moving the firm closer
to the most optimal selections.
Similar to how investment portfolio managers optimize rates of return from an array of
different quality securities (Brealey, Myers, & Allen, 2006), the PCBA manufacturer
must optimize their PCBA portfolio yields based on an array of different quality
components. The only difference between portfolio optimization and PCBA yield
optimization is how to measure the risk. Measuring risk in investments is much easier to
analyze and better understood. Therefore, portfolio managers can choose the appropriate
rates of return for certain levels of risk. However, measuring the risk in component
quality levels is much more difficult. Therefore, the firm must make the decision where
to be on the efficient frontier by appropriately weighting the TCC and the C.
The firm should focus on where the biggest improvements can be realized by working on
the areas where the slope is the steepest on the efficient frontier curve. For example,
moving from Scenario III to Scenario II is much different than moving from Scenario II
to Scenario VI. It costs the firm an extra $85,054 for a 24.78% potential PCBA Yield
improvement by moving from Scenario III to Scenario II. However, by moving from
Scenario II to Scenario VI, the firm will need to spend an additional $394,462 over
Scenario II's TCC for a marginal component yield improvement of 2.0%. Ultimately, the
firm must decide where it wants to be on the efficient frontier and develop an operations
strategy to achieve its goals.
3.5.
Limitationsof the Component Quality Yield Optimization Model
The component quality yield optimization model can be useful for PCBA manufacturing
firms, yet this proof of concept model does come with inherent limitations. First, the
probabilistic model for determining yield fallout calls for a nonlinear program. There are
downsides of nonlinear programs. Professor Gamamik from MIT offers up the following
guidelines for using a nonlinear program19:
Guideline 1: "Unlike linear optimization, nonlinear optimization may be difficult to
solve even with today's computers." This proof of concept analyzes five PCBAs.
However, if this optimization model were to run on a 1000s of PCBAs across hundreds of
different components with five different quality levels, some serious computing power
will be needed.
Guideline 2: "Some nonlinear optimization problems are easy to solve, others are
difficult. The difficulty depends very much on the model's own mathematical structure."
Again, to scale up the model from five PCBAs to over a 1000, the model will need to be
tweaked and refined.
Guideline 3: "Software for solving nonlinear optimization models varies with the degree
of functionality and with the price. The better software packages solve nonlinear models
with many variables and/or constraints." Microsoft Excel will not be powerful enough to
19 Professor
Gamarnik. MIT. Lecture Notes.
scale this optimization model; therefore, a more extensive nonlinear software package
should be used.
Although Microsoft Excel's solver is mediocre at best when solving nonlinear binary
optimization programs, the program was comprehensive enough to run a proof of concept
example for how a component optimization model will work. Overall, the component
quality yield optimization model offers direct insight into how component selection will
impact overall yields and costs for an entire portfolio of PCBAs. The next section
discusses future research that will be valuable in better understanding the true impact of
component selection.
3.6.
Future Research
There is plenty of room for future research on how component quality affects overall
PCBA yields. For one, the complexities of how component to component interactions
affect overall yield is not well understood. Because this model is based on probabilities,
it is very theoretical. Building an empirical model based on hard factory data and
comparing it to the theoretical model would be an ideal approach.
Similarly, incorporating the different cost and quality aspects of moving to new
technologies is another valuable addition to the component quality yield model. There
will always be inherent tradeoffs the firm must make when deciding when to implement
the next best technology. With these decisions, there is more risk since new technologies
may have unknown quality standards. Additionally, newer technologies carry further risk
of market adoption.
Another research topic would be to understand how to effectively push the 60 component
selection process upstream to the design group. Knowing how certain components would
affect yield, designers could incorporate component quality levels into their process to
design better products that would achieve higher yield sooner in the manufacturing
process. Implementing design for six sigma, designing better redundancies and using the
proper component can then be optimized earlier in the process. Moving the process
upstream will keep the designers more intimate with the product and help improve
overall product manufacturing cost and yield performance.
Finally, future research should look to understand how to properly measure the risk of
component selection. Having a keen understanding of risk will allow firms to make
better business decisions of where to be on the efficient frontier.
3.7.
Summary of the Component Yield Optimization Model
As Cisco embarks on a journey to achieve a 60 yield process for every PCBA in their
product portfolio, the firm will undoubtedly need time and resources to achieve its goal.
Cisco has begun the journey by instilling the Six Sigma culture across the firm and
utilizing DMAIC techniques in manufacturing and problem solving. However, only
employing DMAIC on current manufacturing products will allow manufacturing firms to
successfully achieve a 50 process (Bahiuelas & Jiju, 2004, p.251). Realizing the 6c0 goal
requires a product redesign using design for six sigma (DFSS) and IDOV (identify,
design, optimize, and verify) steps (Bafiuelas & Jiju, 2004, p. 2 5 1). Therefore, while
component sourcing engineers and manufacturing managers focuses on improving
current production yields, design engineers should focus on DFSS to ensure more
products hit the 60 yield processes quicker during the product volume ramp up.
The component quality yield optimization model is a way to identify places for yield
improvements by optimizing the component selection across the portfolio. This model
can be used at both the product development phase as well as in current manufacturing.
This is a proof of concept and is meant to explain the procedure for how design
engineers, component sourcing engineers, and manufacturing managers can use this tool
to make better business decisions to ultimately improve every products yield while
lowering the total cost.
The method is based on a nonlinear probabilistic program defined in three main steps:
1. Define and Identify the 6a component quality standard
2. Calculate the Total Cost of the Component
3. Optimize the component selection across the entire PCBA portfolio
Just as investment managers optimize portfolios to maximize return for less risk, PCBA
manufacturers can maximize yield for less cost or choose to spend more for higher yields.
Inherently, there is a tradeoff between minimizing total cost and minimizing yield fallout.
The firm cannot minimize both simultaneously, but it can ensure the component selection
puts the firm on the efficient frontier. Therefore, depending on the firm's operation
strategy, the PCBA manufacturer must decide which approach to employ.
Because of the nature of where the industry resides on the technology S-curve, shopping
for the lowest cost component is not the best approach (Utterback, 1994). I recommend
investing in higher quality components to increase yields and operate closer to Scenario
II to ensure yield fallouts remain as low as possible for the lowest total component cost.
In the end, achieving a 6a yield process will enable so much more for the company
compared to the cost savings gained by negotiating better prices for similar or lower
quality components. Companies should also push designers to incorporate more yield
modeling and design for six sigma when designing products and selecting which
components to use (Kwak & Anbari, 2006). Investing money upfront in the design for
manufacturing process will reap more benefits down the road. Ultimately, this tool
enables employees to make better decisions based on data to move the company in the
right direction to achieve a 60 yield process on every PCBA manufactured.
This page has been intentionally left blank
4. Chapter 4: Hypothesis 2 Results - The PCBA Prioritization
Algorithm using 6ao
Recall Hypothesis 2: Overall PCBA yields will improve by optimally allocating
manufacturing resources to a holisticallyprioritizedlist of PCBAs across the entire
portfolio.
The second hypothesis in this thesis looks at a tool that will assist in improving overall
PCBA yields. The PCBA Prioritization Algorithm was derived using the key steps of a
Six Sigma - DMAIC approach. First, the thesis Defines the current problem, then
Measures the data and Analyzes the results. Next, the thesis looks at how to Improve the
prioritization algorithm, as well as make recommendations on how the prioritization
algorithm can Improve PCBA yields by allocating manufacturing resources to the most
important PCBAs first. Finally, the PCBA prioritization algorithm will provide a means
to Control and monitor all the PCBA yields across the portfolio.
4.1.
Define: The Problem Statement
Right now there are over 1,000 different PCBAs manufactured every quarter at Cisco.
The number of products continues to grow while manufacturing resources are asked to do
more with less. Listen to the voice of the customer, improve yields, meet output demand,
and hit the time to market goals are just a few of the priorities managers deal with day in
and day out. Often times, the burning issues get the most attention and resources are put
to use on what went wrong last week. Instead of investing the time upfront to improve
the capabilities, this type of problem solving leads to a viscous downward cycle where
employees work harder rather than smarter on the wrong issues by using shortcuts
(Repenning and Sterman, 2001, p.73).
There is general need to break the firefighting, reactive work loop and proactively solve
problems through investments to prevent future issues. This is easier said than done, and
as Peter Bigelow explains, "stuff happens [and] all too often unexpected plans or
lengthier-than-expected tasks derail our well-planned, highly ambitious and essential
plans" (Bigelow, 2006, p. 14). The firm needs to be aware of stuff happening and
develop capabilities to refocus efforts on the important business elements: customers,
employees, and suppliers (Bigelow, 2006, p. 14). Managers need to be cognizant that a
change requires an initial setback in time and resources and that the capabilities created
are well worth the investment. This PCBA Prioritization Algorithm is a tool that will
attempt to encourage firms to spend time and resources today to develop the capabilities
to better satisfy customers through increased PCBA yields, while also allowing
employees to work collaboratively, share best practices, and manage suppliers better.
Within the factory network, these thousands of PCBAs are manufactured at any given site
around the world. Local teams ensure their resources are applied appropriately in order
to sustain their process. However, over 110 teams were soon discovered, each managing
their resources to tackle their own issues in the best way they knew how. Seldom did the
teams transfer best practices to other groups. Additionally, these 110 teams were not
managed holistically across the organization. It is not necessarily the team's fault that
they were not communicating best practices; the company just did not have the proper
communication channels to properly link them. Recognizing this gap, Cisco launched an
initiative to allow the proper communication. During the Test Excellence initiative, the
teams voiced their opinions at how to better manage manufacturing resources on poor
yielding PCBAs. Cisco currently prioritizes PCBAs locally rather than holistically.
Therefore, applying a global prioritization algorithm will be Cisco's first unified
approach. Using the information from the change management meetings, working with
Cisco PCBA experts, and benchmarking the industry, this chapter synthesizes the
pertinent information to create a PCBA Priority Algorithm.
The Problem Statement: PCBA yields need to improve and Cisco does not have a
mechanism that defines which PCBAs in the entire portfolio should have priority.
The Vision: The PCBA Priority Algorithm will holistically sort through a list of PCBAs
across the entire manufacturing portfolio and easily determine quickly and correctly
which PCBAs need full manufacturing resources immediately and which PCBAs can be
put on hold for the time being.
At a local level, manufacturing managers may have a strong handle on where and when
to apply resources. However, from a portfolio view, regionally managing the issues may
be a local optimum but not a global optimum. Because there is no portfolio view in
place, it is rather difficult to determine which solutions would be optimal. By looking at
the problem holistically, resources can theoretically be allocated more efficiently and
economically. Best practices can transfer effectively and new capabilities can develop
quickly, and in turn, lead to overall improved PCBA yield performance (Repenning &
Sterman, 2001).
In today's competitive world, it appears manufacturing resources are spread too thin and
asked to do more with less. Lean thinking introduced the concept of eliminating waste to
create opportunities for significant improvement (Womack & Jones, 1996). The PCBA
Prioritization Algorithm shows managers these areas for improvement and provides a
mechanism to allow for waste elimination by allocating manufacturing resources
appropriately. Now that the problem is defined, the next sections will explain how the
data is measured.
4.2.
Measure: The PCBA PrioritizationAlgorithm
The PCBA PrioritizationAlgorithm provides a way to easily measure a large number of
PCBAs from around the globe. The algorithm triages any list of PCBAs and rank orders
them based on multiple manufacturing factors, such as yields, revenue potential, and
customer and management feedback. This tool gives managers the capabilities to
prioritize manufacturing resources to resolve the most pertinent yield and quality issues
for the PCBAs that need the most attention.
The PCBA prioritization algorithm is based on the three main steps. First, it obtains all
the data for the most important manufacturing factors for every PCBA in the list. Next,
using these data as inputs, the second step uses the Technique of ranking Preferences by
Similarity to the Ideal Solution (TOPSIS), a multiple attribute decision making algorithm
(Yoon & Hwang, 1980), to generate a priority list of PCBAs.20 Then, the Total PCBA
Index Score for every PCBA is displayed. The third main step then applies general
priority rules to finalize the priority list. The flow chart in Figure 5 summarizes these
steps.
Figure 5: Three Steps to Determine the Final PCBA Priority List
The PCBA Prioritization Algorithm, Expanded
The PCBA priority algorithm generates a single score than can easily allow managers and
other employees to holistically prioritize thousands of PCBAs manufactured every
quarter. The Total PCBA Index is based on a combination of different best known
prioritization methods already in use and enables these best practices to be shared across
the company.
4.2.1. Scoring each PCBA Based on Manufacturing Data
For every PCBA, the priority algorithm generates one single overall index score
called the Total PCBA Index (TI). The TI is made up of three other index scores than
score various performance factors such as:
1. Revenue & Demand Outlook
2. Quality Performance
3. Customer & Management Perception
20
The TOPSIS model is discussed in full detail in section 4.3.
The Total PCBA Index (TI) is a single score that allows managers to easily determine
every PCBA's performance. The PCBA Priority Algorithm also generates indices
based on the revenue and demand outlook, quality performance, and customer and
management perception. The Revenue & Demand Index (RDI) looks at PCBA
demand over the next quarter and next year while analyzing the volume to cost ratio
to determine its profitability. The RDI also determines if the particular PCBA is
capacity constrained.
Next, the Quality Index (QI) analyzes actual yields compared to a set yield goal.
Additionally, the 13 week yield trend is also analyzed using statistical process control
to determine the process capability of the yields. The third factor in the Quality Index
is the cost of poor quality, or COPQ. This cost measure determines which PCBAs
incur more cost every time there is a factory failure. Similarly, each failure represents
waste in the system, which is also measured. Finally, the cost of the every wasted
unit is recorded.
While the first two indices are quantitative, the final index is based more on
subjective data. The Customer & Management Index (CMI) draws on customer
feedback, which can come in many different communication channels. This index
tries to compile an overall customer experience score. Next, managers determine the
market importance based on internal information. A particular PCBA may be the key
to opening a new market or allowing a new product to be introduced. The fourth
factor in the quality index is determined by the quality engineers. These engineers
are intimately involved with the PCBA's manufacturing process as well as each
PCBA's field performance. Therefore, the quality engineers can forecast if particular
PCBAs will fail, meet, or exceed customer expectations in the future. Finally,
managers determine if a PCBA should be in the Penalty Box. Penalty Box PCBAs
are PCBAs with known issue that cannot be resolved. Management has agreed not to
spend time or money to fix the problem at this point.
Table 23 summarizes how the Revenue & Demand, Quality, and Customer &
Management Indices can be further broken down into several important
manufacturing factors indicated by the following variables: a, 3,y, 6, 8, (, r1,0, 1,K,
and k.
Table 23: Important Manufacturing Factors Used to Calculate Each Index
Important Factors used to calculate the Total PCBA Index (TI)
Revenue & Demand Index (RDI)
a =Next Quarter Demand Forecast
fi =Next Year Demand Forecast
y =Volume / Cost Ratio
Quality Index (QI)
8 =Current 6sigma OR Perfect Yield Delta
c =6sigma OR Perfect Yield Delta 13 week Statistical Trend
=COPQ for not achieving 6sigma or 100% Perfect Yield
t =WASTE: [current 6sigma Yield Delta * Next Qtr Demand]
0 =Ratio [COPQ / Waste] (i.e. $ / Unit of Waste)
Customer & Management Index (CMI)
t =The Customer Experience
K =The Market Importance
. =The Quality Engineers' Expected Performance
Each index is a function of the factors as seen below in Equation 7.
Equation 7: Indices as a Function of the Factors
TI = f (a, 0, y, 6, &,(, q, 0, t, K,X)
RDI = f (a, 0, y)
QI
= f(8, 8, (5,1, 0)
CMI = f(t, K,X)
These factors are inputs into Step 1 of the PCBA Prioritization Algorithm. Data from
each factor for each PCBA are input into the TOPSIS algorithm and then used to
ultimately calculate the different indices, represented by the shaded areas as seen in
Figure 6.
The data from each of the manufacturing factors in Table 23 are fed into Step 1 of the
PCBA Prioritization Algorithm. Data from each factor for each PCBA are then
inputs into the TOPSIS algorithm and used to ultimately calculate the TI, RDI, QI,
and CMI Indices represented by the shaded areas as seen in Figure 6.
Figure 6: PCBA Prioritization Inputs
The PCBA Prioritization Algorithm, Expanded
Step 3:
-p
Apply
General
Priority Rules
Generate
Finalize the Priority
List
The PCBA Total Index ranking system allows upper management to have full
visibility into how the company is performing without knowing too many details
about each PCBAs yields, demand, customer satisfaction, and the like. The PCBA
Total Index quickly points out which PCBAs are the best and which are the worst.
Using quick sorting methods, it is possible to deduce which manufacturing sites are
the best performing and which manufacturing sites need substantial improvements or even which business units develop the best product. The opportunities to slice and
dice the data are endless. Thus, understanding the drivers of these index scores is
necessary in order to make improvements. The following sections detail the three
main indices that contribute to the Total PCBA Index Score: the Revenue & Demand
Index Score, the Quality Index Score, and the Customer & Management Index Score.
4.2.2. The Revenue & Demand Index Score (RDI)
The goal of the Revenue & Demand Index is to identify PCBAs that will generate the
most profit for the upcoming year. This index is based on current manufacturing data
that is readily available. Parameters such as manufacturing capacity, costs, demand,
and profitability can be used to determine the revenue potential of each PCBA. The
Revenue & Demand Index will be scored from 0 - 100. Remember, there are three
main factors that compose the RDI. 2 1 Let's take a closer look at each RDI factor.
4.2.2.1.
RDI Factor 1 & 2 - Demand Forecast (a) and Next Year (a3)
The demand forecast for next quarter allows manufacturing teams to focus on
immediate yield issue that will generate the most revenue at minimal cost. However,
due to the nature of the business, it is also very important to look at a product's
demand forecast over the year to ensure that any work done today will reap the
benefits of yield improvement over a longer time horizon than just one quarter. Since
the demand curves vary from PCBA to PCBA, any number of PCBAs may become
obsolete in the next quarter or not manufactured due to market supply and demand
needs. Therefore, both quarterly and yearly demand curves are needed to paint a
good picture of what volumes manufacturing should expect. Using the overall
demand profile, PCBAs are prioritized appropriately where yield improvements make
the most sense.
4.2.2.2.
RDI Factor 3 y, - Ratio of Volume to Cost
Additionally, it's important to consider manufacturing costs will change over time.
These costs may change because of economies of scale as volumes increase or as
firms move down the learning curve and lower costs by understanding the process
21
Recall from Equation 7, that TI = f(RDI, QI, CMI) and RDI = f(ca, 3, y).
better. To account for these changes in cost, it is necessary to analyze where on the
curve the firm resides for a particular PCBA. To do so, quarter over quarter volume
percentage change versus cost percentage change is calculated. Taking the ratio of
the two can determine if manufacturing costs are increasing, decreasing, or staying
the same versus volume changes from quarter to quarter. This ratio essentially tries
to pinpoint where manufacturing is on the cost curve. PCBAs with higher costs
compared to their volume change generate less revenue at more cost and should be
prioritized higher. Whereas the products with lower costs and increasing volumes are
realizing significant economies of scale and learning benefits as the manufacturing
costs become cheaper and cheaper.
4.2.2.3.
Putting the Revenue & Demand Index together
Using a the TOPSIS algorithm with inputs from upcoming quarterly and annual
demand forecasts and cost data, each PCBA is given a Revenue & Demand Index
Score (see Figure 7). Remember that TI = f(RDI, QI, CMI) = f(a, 0, y, 5, ý, r, 0,
0, 1,
K,
X), where RDI = f(a, 1, y).
Figure 7: Calculating the RDI
The PCBA Prioritization Algorithm, Expanded
An index score of 100 indicates the best Revenue & Demand PCBA for the firm. The
upcoming quarter and annual demand is very healthy. Manufacturing costs continue
to be reduced with either increasing or steady volumes. Additionally, gross margins
may be increasing. In general, PCBAs with a higher rating have the highest revenue
generating potential. The opposite is true for PCBAs with an index score of 0. These
low scoring PCBAs may have no future demand, increasing manufacturing costs, or
decreasing gross margins.
Overall, the goal of the Revenue & Demand Index is to identify which PCBAs in the
portfolio have the most revenue producing potential. Therefore, PCBAs that have the
highest positive future demand and most potential to reduce costs by quickly moving
down the cost curves should be prioritized first.
4.2.3. The Quality Index Score (QI)
The quality index is also based on a score from 0 to 100. The goal of the quality
index is to filter PCBAs with low and worsening yields in addition to a high cost of
poor quality. Cost of poor quality, or COPQ, is the total cost associated with any
quality issues that may arise in the manufacturing process. These costs can be caused
by additional labor, debug, repair, equipment usage, and time to market factors.
Higher quality index numbers correspond to PCBAs with better quality. The quality
index is calculated based on manufacturing data derived from five factors
manufacturing factors. 22 Let's take a closer look at each QI factor.
4.2.3.1.
QI Factor 1 - 5, The 6o OR Perfect Yield Delta
Using a proprietary yield calculation, the manufacturing firm can determine a 6a
yield goal for every PCBA. Then, the actual yield of the PCBA is compared to its 60
goal and the resulting difference is called the 6ayield delta. The actual yield for the
PCBA is also compared to perfect yield, or 100%. This delta is called the perfect
yield delta. In both cases, the larger the delta, the further the PCBA is from its yield
goal. The quality index focuses first on the 60 yield delta and then on the perfect
yield delta. Because some PCBAs may yield above their respective 60 yield goals, it
is necessary to focus on both the 60 and perfect yield deltas. This allows the
algorithm to proactively catch PCBAs with yields greater than their 60 goal that are
trending lower.
The greater the 60 yield delta, the greater the possible yield improvement. There are
diminishing returns as the 60 yield delta approaches zero as the same time, money,
and resources needed to improve the yield produces a lower and lower yield
improvement.
22 Remember
that TI = f (RDI, QI, CMI) = f(a, 3, y, 8, 6, r,71, 0,
, K,
X), where QI = f (8,
e, ý, i,
0).
4.2.3.2.
QI Factor 2 - E, The 60 OR Perfect Yield Delta 13 week Trend
Additionally, the 60 yield and perfect yield deltas are tracked using statistical process
control. Therefore, the process capability of each PCBA can be determined. More
manufacturing resources should focus on PCBAs with yields trending lower. If
PCBA yields are trending higher, best practices should be shared to help improve
PCBAs with worse yields. The quality index accounts for the 60 and perfect yield
deltas along with the past 13 week trends to determine how the PCBAs should be
prioritized in order to maximize yields.
4.2.3.3.
QI Factor 3 - C, The Cost of Poor Quality (COPQ)
Every time a PCBA fails a test, a cost is incurred. This cost of poor quality includes
retest, scrap, defect escapes, repairs, false rejects, diagnostics, programming,
maintenance, equipment, and labor costs (Michel & Reinosa, 2004). If a PCBA is
yielding at its 60 goals, this means that the COPQ number is at an acceptable level;
however, in the spirit of continuous improvement the 60 goal should be recalculated.
PCBAs with a high COPQ should be flagged by the quality index since these PCBAs
are costing the firm more money. If the quality issues can be resolved sooner rather
than later, the company will realize the cost savings quicker. Manufacturing
resources should understand why these costs are so high and work to bring the costs
down. The COPQ incurred will be used with the amount of waste produced, as
described below, to determine how much money is spent on the poor yields. The goal
of the quality index will be to find PCBAs with the highest COPQ and most Waste
and then prioritize them higher.
4.2.3.4.
QI Factor 4 - r7 , Eliminating the Waste in the System
Every time a PCBA fails a test, wasted efforts are created in the system. And each
non value added step incurs a cost that the customer is not willing to pay. Therefore,
the goal should be to eliminate all the waste and its associated costs. One way to
predict the waste in the system is to take last quarter's actual yield and multiply it by
next quarter's volume forecast. This number gives an approximation of how much
material waste to expect in the system. PCBAs with a high amount of waste should
be prioritized higher.
4.2.3.5.
Q/ Factor 5 - 0, The Ratio of COPQ to Waste
The ratio of COPQ to waste will effectively determine how much money is spent on
each PCBA that fails a test step. Higher ratios coincide with poorer performing
PCBAs. Therefore, the quality index places priority on these PCBAs.
4.2.3.6.
Other Costs to Consider
Depending on the firm, there may be other factors and costs to use when determine
the prioritization level of a particular PCBA. For instance, looking at the cost of time
to market, field failures, or new test fixtures may prove to be helpful when
determining the PCBA priority (Michel & Reinosa, 2004). For this thesis, these costs
are out of scope but their effects should be studied in future research.
4.2.3.7.
Putting the Quality Index together
Using a the TOPSIS algorithm with inputs from yield, cost of poor quality, and waste,
each PCBA is given a Quality Index Score (see Figure 8).
INEE
Figure 8: Calculating the QI
The PCBA Prioritization Algorithm, Expanded
An index score of 100 indicates the best Quality PCBA the firm manufactures. In
general, PCBAs with a higher scores have higher quality performance. The opposite
is true for PCBAs with quality index scores closer to 0. These lower scoring PCBAs
are yielding very poorly with increasing waste and cost to fix the yield issues.
Overall, the goal of the Quality Index is to identify which PCBAs in the portfolio
have the worst quality. These PCBAs with poor yields, high costs, and increased
waste will be given a lower Quality Index score and prioritized first.
4.2.4. The Customer & Management Index Score (CMI)
The Customer & Management Index is the last index and also ranges from 0 to 100.
A higher score corresponds to PCBAs that are performing well according to
customers and management expectations. This score is the only subjective score in
the PCBA Priority Algorithm since it is based mainly on perceived performance.
Quality engineers are responsible for assigning the appropriate score based on several
management factors and customer inputs such as field failures, immediate returns,
and surveys. Remember, there are three main factors that make up the CMI.23 Let's
take a closer look at each CMI factor.
4.2.4.1.
CMI Factor I - i , The Customer Experience
The first factor for the customer and management index is based on direct customer
feedback. This factor identifies how the customer perceives the product's
performance. These customer experiences can range from extremely great to
extremely disappointing. Customer perception data seeps into the organization via
many different communication channels, including phone calls, emails, immediate
returns, dead on arrivals products, and customer surveys. This data needs to be
consolidated across the different organization within the firm. Quality engineers
provide one link between the different business units and the problems and can work
to aggregate the data in order to determine which PCBAs are doing great and which
are performing poorly. Other departments such as customer support and field returns
also get various data regarding the customer experience. The CEO or various VPs or
directors may also get direct customer feedback. In any of these cases, compiling
data to determine the customer experience level will be challenging. With the proper
information, the quality engineer can assign a customer experience value to the
specific PCBA. This algorithm uses a high, medium, and low ranking for the
customer experience. Here, a high customer experience value means the customer is
very pleased with the product's performance.
4.2.4.2.
CMI Factor 2 - K, The Market Importance
The market importance factor is intended to give management a voice in the
prioritization process. This voice allows management to reprioritize PCBAs under
certain market conditions. For instance, if a particular PCBA has opportunities to
open new markets or allow a release of a new product, managers can change the
market importance factor. In the same vein, if a manager knows a PCBA will soon be
23
Remember that TI = f(RDI, QI, CMI) = f(a,
3, y, 6, c, ý, rl, 0, t, K,X), where CMI = f(t, K,k).
replaced by a new version, then the market importance factor should also change, as
this PCBA will soon become obsolete. Additionally, managers may have visibility
into the pricing and cost structure of their products. Therefore, as gross margins
change, so to will the market importance factor. Market importance factors will
range from very high, high, medium, low, or very low. By changing the market
importance factor, PCBAs can be reprioritized to allow manufacturing resources to
work on the most crucial issues for the most important PCBAs first.
4.2.4.3.
CMI Factor 3 - A, Quality Engineers' Expected Performance
Quality engineers work with the customers to develop a customer experience score as
already described. If the customers are experience particular problems with a certain
PCBA, quality engineers have the ability to understand the customer experience.
Quality engineers and managers also have data of the historical interactions with the
customer and future forecasts for product quality and reliability in the field. With this
data, the quality engineers can assign a score of high, medium, or low as to how they
perceive the PCBAs will perform in the future compared to management and
customer expectations.
4.2.4.4.
Putting the Management and Customer Index together
The Management and Customer Index is a subjective score based on three main
factors: (1) the Customer Experience, (2) the Market Importance, and (3) the Quality
Engineers' Expected Performance (see Figure 9).
Figure 9: Calculating the CMI
The PCBA Prioritization Algorithm, Expanded
Analyzing how the customer perceives each specific PCBA's performance allows the
manufacturing team to make better business decisions about which problems to
resolve first. Understanding how the PCBA will affect new markets or products is
essential in determining which PCBAs are more important than others. Finally, a
quality engineer works with these products and their customers on a day-to-day basis,
has historical data, and knows management's expectations. Quality engineers add
valuable knowledge to each PCBA. Tying these factors together allows every PCBA
to attain a single Customer & Management Index score. This score is then weighted
accordingly and used in the PCBA Total Index calculation.
This section concludes the Measurement portion of the Six Sigma DMAIC process.
All the important manufacturing factors have been obtained from different databases
and categorized under different indices. Next, in order to convert raw manufacturing
data into index scores, it is necessary to Analyze the manufacturing factor's data. The
next section will describe the analysis based on TOPSIS algorithm.
4.3.
Analyze: Putting the PCBA Total Index Together using TOPSIS
The section focuses on Step 2 of the PCBA Prioritization Algorithm and how the
Technique for Order Preferences by Similarity to Ideal Solutions (TOPSIS) algorithm
will calculate each index. Using the raw data from all the important factors in step 1, the
PCBA Algorithm then applies the TOPIS technique to generate the Total PCBA Index.
Additionally, the algorithm also calculates the Revenue & Demand Index, the Quality
Index, and the Customer & Management Index. The TOPSIS algorithm is based on the
following steps (Franklin & Niemeier, 1998, p2 2 - 26), (Olson, 2003, p722):
*
TOPSIS-Step 1: Obtain PCBA Raw Scores & Determine the Ideal States
*
TOPSIS-Step 2: Normalize the Raw Scores
*
TOPSIS-Step 3: Weight the Normalized Scores
*
TOPSIS-Step 4: Determine the Priority Index based on the Ideal States
*
TOPSIS-Step 5: Display the Index Score
TOPSIS will prioritization the highest revenue generating, lowest quality, and lowest
customer and management performing PCBAs. The following sections describe each
TOPSIS step in full detail.
4.3.1. TOPSIS-Step 1: Obtain PCBA Raw Scores & Determine the
Ideal States
The first step in the TOPSIS algorithm will compile all the raw data obtained for each
PCBA for each of the given factors discussed earlier. The raw data for these factors
contribute to how the PCBAs will ultimately be prioritized. In order to determine
which PCBA deserves more priority than another, each PCBA must be compared to
an ideal situation. Therefore, the TOPSIS algorithm requires each PCBA be
compared against a best and worst case state, referred to as Ideal Positive, (f), and
Ideal Negative, (T) states (Franklin & Niemeier, 1998).
The Ideal Positive state is the best case situation, or, in other words, the best
performing PCBA. Thus, this best case PCBA will have a very high Revenue &
Demand Index score, a very high Quality Index score, and a very high Customer &
Management Index Score, and therefore, be prioritized the lowest.
Similarly, the Ideal Negative state is the worst case situation. This board will have a
very high Revenue & Demand Index score, but a very low Quality Index score, and a
very low Customer & Management Index Score. This worst case PCBA will be
prioritized the highest. Both the Ideal Positive and Ideal Negative states can be seen
in Table 24.
Table 24: Ideal States
Revenue & Demand (RDI)
Quality (QI)
a
Customer & Management (CMI)
a~ r
0~
a
a
0%
060
0%
-a
"a
GO
EE
ca
o
o
.a
a
C
0.
Ideal State
Ideal Positive. I"
Ideal Negative, FI
No
Yes
2
a,
P
a
a
2
a
a
a
a
0%
a > 0%
.•
a,
2
9z
0%
r•
a
t
C
a
a
a
a
ae
-
0
a
a
100%
5
a
100%
100%
C.
o
100%
._•
ar~
100%
b:
ud2
a
a
CL
aS
a
a
C.
2u
a
a
a
~
0
*U0
a
a
'
Z
0%
Z
0%
2
0%
100%
100%
100%
100%
100%
100%
100%
100%
0%
0%
0%
0%
0%
.4
High
Low
.o
aa
a
-l
a
a
a
Very Low Exceeds
Very High
Fails
No
No
No
When ranking the PCBAs, the Ideal Positive state will get the highest Total PCBA
Index score and the lowest priority. Similarly, the Ideal Negative will receive the
lowest Total PCBA Index score and the highest priority. Using the factors outlined in
section 4.2 (a, 0, y, 8, c, , r5, 0, t, K, k), the ideal positive and negative states can be
determined (See Table 24).
To appropriately determine each index score, the raw score for each factor needs to
be normalized. Normalizing the raw data eliminates any units so each score becomes
dimensionless (Franklin & Niemeier, 1998). By having dimensionless numbers, each
factor can then be weighted appropriately, and each PCBA's Total Index score
compared against the IdealPositive and Negative states to determine its relative
position in the overall priority list.
4.3.2. TOPSIS Step 2: Normalize the Raw Scores
Each raw score for each factor for each PCBA is divided by the root-sum-of-squares
of the factor to determine the normalized value (Franklin & Niemeier, 1998). For a
given PCBA, Equation 8 shows how each factor's raw score, Xi, is converted into a
normalized value, Xi*.
Equation 8: Normalizing Each Factor's Score
X
=i
P represents the number of PCBAs being prioritized.
For example, based on a set of raw manufacturing data, normalized values for each
factor are calculated for ten randomly selected PCBAs. 24 The normalized scores are
denoted by the * as seen in Table 25. Each factor is listed under its appropriate index,
and the Ideal Positive and Negative states are also shown.
24
All PCBA data is disguised to protect Cisco confidentiality.
Table 25: Ten PCBA Normalized Scores for Each Index
Step 2: Normalize the Raw Scores
PCBA #1
PCBA #2
PCBA #3
PCBA #4
PCBA #5
PCBA #6
PCBA #7
PCBA #8
PCBA #9
PCBA #10
0*
0.34
0.34
0.34
0.29
0.29
0.27
0.24
0.29
0.19
0.34
y***
0.35
0.31
0.31
0.36
0.34
0.25
0.26
0.35
0.23
0.02
CMI
QI
RDI
PCBA & Ideal
State Scores
0.33
0.33
0.21
0.35
0.31
0.33
0.34
0.22
0.35
0.02
8*
0.02
0.02
0.02
0.01
0.03
0.07
0.26
0.43
0.25
0.34
C*
0.02
0.05
0.05
0.01
0.36
0.31
0.15
0.13
0.15
0.20
0.01
0.09
0.09
0.03
0.03
0.19
0.26
0.09
0.09
0.30
1*
1*
0.00
0.00
0.00
0.02
0.02
0.27
0.31
0.27
0.14
0.12
0*
1*
K*
X*
0.07
0.07
0.07
0.29
0.29
0.20
0.05
0.27
0.20
0.18
0.38
0.38
0.38
0.38
0.38
0.13
0.25
0.13
0.25
0.13
0.10
0.10
0.10
0.20
0.20
0.10
0.20
0.20
0.49
0.49
0.38
0.38
0.38
0.25
0.25
0.38
0.13
0.13
0.25
0.25
Ideal Positive, I+ 0.00 0.00 0.00 0.43 0.41 0.39 0.35 0.30 0.13 0.49 0.13
Ideal Negative, I
0.34
0.37
0.36
0.00
0.00
0.00
0.00
0.00
0.38
0.10
0.38
4.3.3. TOPSIS Step 3: Weight the Normalized Scores
After normalizing the raw score, the yield engineer will weight the different factors
based on their relative importance (Franklin & Niemeier, 1998). The weighted scores
are calculated by multiplying the normalized scores in Table 25 by the respective
weightings. The weighted scores are denoted by X&* and the results are seen in Table
26. This example gives equal weightings for each index. Section 4.4.1 details the
weightings further by looking at a sensitivity analysis of the rankings based on
different weighting scenarios.
Table 26: Ten PCBA Weighted Scores for Each Index
Step 3: Weight the Normalized Scores
CMI
QI
RDI
PCBA & Ideal
State Scores
a*w
P*w
Y*w
6*w
E*w w*,*
q*W
0*w
t*w
K*w
3*,
PCBA #1
PCBA #2
PCBA #3
PCBA #4
PCBA #5
PCBA #6
PCBA #7
PCBA #8
PCBA #9
PCBA #10
3.79
3.79
3.79
3.24
3.17
2.95
2.69
3.25
2.09
3.80
3.90
3.39
3.39
4.02
3.77
2.78
2.94
3.89
2.55
0.17
3.64
3.64
2.37
3.84
3.48
3.69
3.78
2.48
3.84
0.28
0.14
0.14
0.14
0.06
0.20
0.48
1.73
2.85
1.68
2.30
0.14
0.33
0.33
0.08
2.43
2.04
0.98
0.84
0.97
1.36
0.05
0.60
0.60
0.21
0.21
1.27
1.77
0.57
0.57
2.03
0.02
0.02
0.02
0.12
0.12
1.79
2.06
1.79
0.92
0.83
0.47
0.47
0.47
1.95
1.95
1.32
0.31
1.79
1.36
1.23
4.17
4.17
4.17
4.17
4.17
1.39
2.78
1.39
2.78
1.39
1.09
1.09
1.09
2.19
2.19
1.09
2.19
2.19
5.47
5.47
4.17
4.17
4.17
2.78
2.78
4.17
1.39
1.39
2.78
2.78
Ideal Positive, I+
Ideal Negative, I
0.00
3.82
0.00
4.10
0.00
3.95
2.88
0.00
2.73
0.00
2.61
0.00
2.32
0.00
1.99
0.00
1.39
4.17
5.47
1.09
1.39
4.17
4.3.4. TOPSIS Step 4: Determine the Priority Index Based on the
Ideal States
With the weighted scores, the PriorityIndex (PI)can now be calculated. TOPSIS
calculates the priority index by comparing the weighted scores to the ideal state. This
is completed by taking the distance every PCBA is from both the ideal negative and
idealpositive states, and then, determining which PCBA is closest to the ideal
negative state. D - and D + will represent the distance the PCBA's weighted score is
from the ideal negative and ideal positive states, respectively. Equation 9 and
Equation 10 show how D- and D + are calculated. The priority index is defined in
Equation 11.
Equation 9: Calculating D
i=
(
X,*= The normalized weighted score for the ith factor
(I>,i
factors's Ideal Negative State
= The normalized weighted score for the ith
Equation 10: Calculating D +
th
X,. = The normalized weighted score for the i factor
(I0••)
= The normalized weighted score for the it factor's Ideal Positive State
Equation 11: Calculating the Priority Index
PI =
DD ÷ +D-
In the example in this section, based on the weighted scores in Table 26, Table 27
shows the D+, D -, and priority index (PI) for each index across the 10 PCBAs.
Notice how the ideal positive and negative state priority indices are 1 and 0,
respectively, for all the indices.
Table 27: Priority Index for Each Index
Step 4: Determine Priority Index Based on the Ideal States
D+
CMI
D-
PI
D+
TI
D-
PI
5.88
5.88
5.88
4.52
4.52
5.19
3.57
3.28
1.96
1.39
0.00
0.00
0.00
1.77
1.77
2.78
3.29
4.08
4.80
5.37
0.00
0.00
0.00
0.28
0.28
0.35
0.48
0.55
0.71
0.79
10.28
9.93
9.54
9.35
8.65
8.11
7.13
7.12
6.36
4.67
0.64
1.15
1.93
2.71
3.70
4.61
4.99
5.89
5.94
8.44
0.06
0.10
0.17
0.22
0.30
0.36
0.41
0.45
0.48
0.64
1.00
0.00 5.65 1.00 0.00
5.88
1.00
0.00
10.65
1.00
0.00
5.65
0.00
0.00
10.65
0.00
0.00
D+
RDI
D-
PI
D+
QI
D-
PCBA #1
PCBA #2
PCBA#3
PCBA #4
PCBA #5
PCBA#6
PCBA #7
PCBA#8
PCBA #9
PCBA#10
6.54
6.25
5.61
6.43
6.03
5.49
5.49
5.65
5.06
3.81
0.38
0.78
1.73
0.59
0.86
1.60
1.63
1.59
2.33
5.39
0.05
0.11
0.24
0.08
0.13
0.23
0.23
0.22
0.32
0.59
5.33
4.99
4.99
5.06
4.23
2.96
2.83
2.84
3.32
2.31
0.51 0.09
0.84 0.14
0.84 0.14
1.97 0.28
3.13 0.43
3.31 0.53
3.38 0.54
3.94 0.58
2.60 0.44
3.67 0.61
Ideal Positive, 1+
0.00
6.86
Ideal Negative, IF
6.86
0.00
PCBA & Ideal
State Scores
PI
0.00 0.00
5.88
With the priority indices calculated, the PCBA Priority Algorithm's next step will be
to convert the priority index into a score between 0 and 100 so users will be able to
easily decipher the scores.
4.3.5. TOPSIS Step 5: Display the Index Scores
The PCBA Prioritization Algorithm, Expanded
Step 1:
Obtain all PCBA's
Data for Important
Factors
Step2:
Apply the TOPSIS
Apply the TOPSIS
Algorithm
Step 3:
-
Generate
SApply General -- ' Finalize the Priority
Priority Rules
List
Using these priority index scores, the Total PCBA, Revenue & Demand, Quality, and
Customer & Management Index scores are calculated and ranked. Now, the first two
steps in the PCBA Prioritization Algorithm are complete and the Total PCBA Index
Score, or TI, can be displayed as seen in Table 28.
Table 28: Index Scores Based on the Priority Index
Step 5:
Prioritize
PCBA #1
PCBA #2
PCBA #3
PCBA #4
PCBA #5
PCBA#6
PCBA #7
PCBA #8
PCBA#9
PCBA #10
TI
RDI
QI
CMI
Sccore Sccore Sccore Sccore
6
95
9
0
10
89
14
0
17
76
14
0
92
22
28
28
30
87
43
28
36
77
53
35
41
77
54
48
45
78
58
55
48
68
44
71
64
41
61
79
Ideal Positive, I +
100
0
100
100
Ideal Negative, F
0
100
0
0
TOPSIS generates a priority list with unique index scores that are ranked in priority
order according to the manufacturing factors for each PCBA. The next step in the
PCBA Priority Algorithm is to apply the manufacturing firm's general priority rules.
4.3.6. Apply General Priority Rules to the Total PCBA Index
The PCBA Prioritization Algorithm, Expanded
In the third and final step of the PCBA prioritization process, the PCBA Prioritization
Algorithm overlays four general category rules mandated by the firm to determine the
final priority ranking as seen in Figure 10. These rules are a final attempt to tweak
the priority list after the TOPSIS algorithm to ensure manufacturing resources are
allocated to the most important PCBAs first.
Figure 10: General Priority Rules
General Priority Rankings Ruels for the
Total PCBA Index
Priority 1:
Capacity Constrained PCBAs
Priority 2:
Regular PCBAs
Priority 3:
No Demand PCBAs
Priority 4:
Penalty Box PCBAs
First, any Capacity Constrained(CC) PCBAs are ranked the highest on the priority
list. Theory of constraints tells us that any time lost in the capacity limiting operation
is time lost forever (Goldratt & Cox, 1992). Here, time equals money, and the
revenue that could be generated in a capacity constrained situation will be lost
forever. Thus, and capacity constrained PCBAs should be worked on first.
Second, any PCBAs with known issues that management has deemed unfixable at
this moment in time will be placed in its own category and placed at the bottom of the
priority list. Because of the nature of these particular PCBAs, these PCBAs are
placed in the Penalty Box (PB). Management may know and understand that a
certain PCBA will never be able to reach its full yield performance potential due to an
inherent design issue, an obsolete or highly defective component, or other known
defect issues. Because of the PCBA's bad performance, like in hockey, these PCBAs
are put into the penalty box for a limited amount of time until it is deemed reasonable
that further action is needed.
Third, any PCBA with No Demand (ND) in the future should be ranked lower on the
priority list, just above the penalty box. It is not worthwhile to focus manufacturing
resources on products that soon will not be made.
Finally, the bulk of the PCBAs will fall into the Regular PCBA category. These
PCBAs are not capacity constrained, have no known defect issues, and have future
demand. These are PCBAs that manufacturing can analyze to determine which need
more manufacturing resources and which do not.
These four general rules tie into the PCBA Prioritization algorithm to generate the
final priority list as see in Figure 11.
100
Figure 11: Four General Rules Tie into the Final Priority List
The PCBA Prioritization Algorithm, Expanded
Generate
Finalize the Priority
List
K
(1, P,y,,ka, ,;,q, 0, K,
9X)
Customer
&
Management
Index (CMI)
f(i, K,1)
In short, the PCBA Priority Algorithm uses a final set of priority rules after the
TOPSIS algorithm completes to determine the final rank order of the Total PCBA
Index scores.
4.3.7. Generate Final Priority Rankings on the Total PCBA Index
The PCBA Prioritization Algorithm, Expanded
The finalized priority list of PCBAs is the output of the PCBA Priority Algorithm.
The index scores for every PCBA will range from 0 to 100. A PCBA with a lower
Total PCBA Index score of 0 needs immediate attention and manufacturing resources
to resolve manufacturing performance issues. In contrast, a PCBA with an index
101
..•-•,-is
IColumn2
Ldar.d
Tetalm
Ta4q,
Column3
• x •*='*
• l)tal•/aul*x
•'•Column4
•, I0IoweJt
- tO01
• Column5
SereTotal]bdex
IREII•
Column6
I0- IW1 Column7 Column8
&
.R•-t•ae
score of 100 is meeting or exceeding all its metrics for each criterion. This board
needs no additional manufacturing resources at this time, but should always be
monitored using statistical process control to ensure its good performance does not
deteriorate. Additionally, with continuous improvement in mind, the goals for these
over performing PCBAs ought to be reevaluated. With limited manufacturing
resources, it is imperative to work on the most important PCBAs first. Therefore, the
prioritization algorithm places a higher priority on PCBAs with the lowest Total
PCBA Index Score, and Table 29 details how the PCBA Total Index will look,
allowing a manager to prioritize a list of PCBAs to ensure the most important issues
are resolved in the most appropriate and timely manner.
Table 29: The PCBA Total Index with Comments
P(*]BA
Tetd
Y,
ndez Scetbq[
Sysm
Scaing •J•m
TetdlIdex
PCBA
•i•,,n
Imex
withsh
Colmm•l
Hilboxta
CoAum1 Column2 Column3 Column4 Column6Column6 Column7 ColumnS
A
implaem PCBd
Ttal
PCBA
lowest
I•']BA.
PCBA
I Total tIe See t0 -100I
(
RIve.
d
&Dma
IDU
0 - 1001
lldexStoen
The PCBA Total Index makes it simple for managers to easily determine which
PCBAs perform better or worse than others do across the entire portfolio. The PCBA
Prioritization Algorithm compiles data for each manufacturing factor from many
different databases and synthesizes the data into a single PCBA Total Index score for
each PCBA in the list. Each row in Table 29 represents a different PCBA, here
102
labeled PCBA#1 to PCBA#35. This table displays the priority list for this particular
set of 35 PCBAs. Manufacturing resources would first be allocated to PCBA#1,
followed by PCBA#2, #3, #4 and so on.
The second column represents the PCBA Total Index score, ranging from 0 to 100.
The algorithm prioritizes PCBAs with the lowest score first.
The third, fourth, and fifth column represent the Revenue & Demand Index (RDI), the
Quality Index (QI) and Management & Customer Index (CMI) scores, respectively.
These three index scores also range from 0 to 100, where 0 indicates the worst score
and 100 indicates a perfect score for that particular index.
The RDI identifies the revenue generating potential of a particular PCBA based on
factors such as upcoming demand forecasts and cost structures. A score of 0
indicates there is poor revenue generating potential. An RDI score of 100 indicates a
very high revenue generating potential in the upcoming year (see section 4.2.2).
The QI ranks PCBAs from worst to best in terms of quality standards and is based on
several factors such as yields, cost of poor quality, and waste generated (see section
4.2.3). A score of 100 indicates the PCBA is meeting all its goals while a score of 0
indicates the particular PCBA's quality levels are very poor.
The CMI ranks PCBAs based on how the Customer and Management perceive the
PCBA's performance to be. A high index score of 100 indicates that the PCBA meets
the customer and management expectations while an index score of 0 suggests the
PCBA could improve its perceived performance (see section 4.2.4).
The manufacturing location (Mfg Loc.) and business unit (BU) for each PCBA can be
seen in columns six and seven, respectively. These columns allow a manager to
further dissect problems that may be associated with lower performing PCBAs. If,
for example, most of the lower performing PCBAs originate from the same
103
manufacturing site, this column will allow the user to identify this problem. Once the
issue is known, an action plan can be constructed to implement a proper resolution.
The eighth column is used for comments on the specific PCBA. As seen in Table
29's comments column, the algorithm buckets several PCBAs into the firms general
priority rules and labels them appropriately: Capacity Constrained,No Demand, or
Penalty Box. Capacity Constrained(CC) PCBAs are ranked the highest on the
priority list. Penalty Box (PB) are labeled with a "PB-" prefix and prioritized in the
penalty box using the Total PCBA Index, indicated by the number displayed. No
Demand (ND) PCBAs are labeled with an "ND-" prefix. Finally, the bulk of the
PCBAs will fall into the Regular category. These PCBAs are not capacity
constrained, have no known defect issues, and have future demand. These are
PCBAs that manufacturing can analyze to determine which need more manufacturing
resources and which do not.
4.3.8. Validate using 27 Extreme Corner Case Scenarios
To ensure the PCBA Prioritization Algorithm will properly prioritize all different
types of PCBAs, this section validates the most extreme comer cases. Based on the
statistical method to generate multiple combinations for a given set of factors, a
design of experiments (DOE) matrix was generated as seen in Table 30. To test all
the extreme cases, the (1) Revenue & Demand, (2) Quality, and (3) Customer &
Management Indices are varied by three different levels. These levels are indicated
by the "+," "0," and "-" symbols representing high, medium, and low.
Table 30: DOE Matrix
#
RDI
QI
CMI
#
RDI
QI
CMI
#
RDI
QI
CMI
1
+
+
+
+
+
+
+
+
+
+
+
+
+
10
11
12
13
14
15
0
0
0
+
16
0
0
+
17
0
0
-
0
18
0
0
0
+
+
+
0
0
0
+
0
+
0
+
-
19
20
21
22
23
24
25
26
27
-
0
0
+
+
+
-
2
3
4
5
6
7
8
9
-
0
0
0
-
0
+
0
+
0
104
0
+
0
-
-
0
Thus, the PCBA Priority Algorithm analyzed these 27 different PCBAs as seen in
Table 31.
Table 31: DOE for 27 PCBAs
of
PCBA TotalIndex Scori(g Svem for 27 DiffeeanrCombmnations
Revenue & Demand(RDI)I. Quality (QI). andCustomerI&Management(0C,') Index Sooare
Hafhestpaoaityis platedon PCBAswith the lowest PCBA Total Index Scotr
Scoe 10- 1011
PC'BATotal aIndex
Revenue & )emandIndex Sora 10- 1001
I
I II
I OsabrIndex
o•st
-1001
0
I
I0- 1001
Custmer & ManazemenrIndex Score
I
I
(RDI. OL CMII
.01
A2
A3
.04
AW
A5
(+,-.
0)
21
(+0.-)
(+,0'.0)
21
30
A-)
32
(.+)
+.-)
(0..0)
35
36
40
#10
(0.0'-)
(+01+)
40
41
0ll2
#12
(++,+)
.6
47
.48
.#9
(+.+.0)
42
(0., +2~
(0.0.0)
(0.+.-)
(0. 0+)
CO.
+,0)
50
50
50
60
60
48
100 0
100 0
100 500
50
50 0
100 0
100 100
50
50
100
100
100
100
lowQI &CMI scores.thisPCBAneedsfullresources
andDemandoutlookCoupledwithle
0 VeryhealthyRevenue
help.
andDemandoutlook.Witha low QI &averageCMI scores, ths PCBA needs
50 VeryhealthyRevenue
help
andDemand
outlookWitha lowCMI & averageQI scores.ttesPCBA needs
00 Veryheahy Revenue
VeryhealthyRevenue
andDemandoulook Witha averageQI & CMI scos. thisPCBA is hiher onthe pnoritylistthanotherPCBAs.
lte
RevenueandDemandoullooka weak
a therankrgs eventhogh
0 VerypoorQI andCMI scores
pushthisPCBAhagher
attentionimprovingQualitysues
100 Withhelthy Rgveme andDemandoutlook.thisPCBA needs
issues
attentionunprovis Customer&Management
0 WithhealthyRevenueandDemandoutlook.thisPCBAneeds
VerypoorQI sorepushest PCBAhigherin therankingseventhoughtheRevenueandDemandoutlookis weak
0 Akhoughaerage ualty scoresandlow Customer
and Demandpushthis PCBA'spnoty lower
& Managementscores,mediocreRevenue
100 AgoodperformngPCBAdoesneed to by hi stonthelist But therestlneedsto befocuson mproving
Qualityscores.
50 Goodperformag PCBA Need tofocuson suproovgCustomer
&Management
scores
100 ThisPCBA has thebest RDI,Q, &0MI scores andthereforeis oneofthe lowerPCBAs onthe list Conrmuetomortor andimprove.
outlook,
and,therefore,thesePCBAs rerankedlower
ThefolowrinPCBAs(#13 - #18) havea weakRevenueandDemand
higher.Meanwhile, maymake
thanPCBAswithhealthydemand f demandshouldpickup,thesePCBAswilget pnoniomed
moresense
toanprove
PCBA wdh a ligherRevenueandDemandscore
50
0 50
5050
100
100
The results from the DOE table are promising. In fact, all the extreme cases are
prioritized logically and correctly. Therefore, for the most extreme corner case, the
PCBA Priority Algorithm is valid. Any other combination for the three indices will
fall somewhere in between these comer cases in the correct priority.
4.3.9. Summary of the PCBA Prioritization Algorithm
In summary, the PCBA Prioritization Algorithm provides a tool to Analyze the raw
manufacturing data to generate a priority list of PCBAs. Several index scores are
calculated to determine the revenue potential, quality levels, and customer and
management perception. Most importantly, the Total PCBA Index scores give users
an easy way to quickly determine the manufacturing health of each particular PCBA
in the firm's portfolio. With limited manufacturing resources, PCBAs with the most
revenue potential, the lowest and worsening quality levels, and the lowest customer
and management expectations receive the highest priority. Manufacturing resources
105
can be reallocated and optimized across the firm as needed to tackle these important
issues. By standardizing the prioritization algorithm across the entire company, the
firm will ensure its resources are efficiently working on the most vital issues at hand
while leaving no revenue on the table.
The TOPSIS algorithm is a crucial piece of the PCBA Prioritization Algorithm. It
ties the raw data obtained from important manufacturing factors and converts these
data into index scores. The Analysis of DMAIC is now complete. The next section
will not only detail how to Improve the prioritization algorithm, but also, more
importantly, show how to start Improving overall PCBA yields today!
4.4.
Improve: The PCBA PrioritizationAlgorithm and Overall PCBA
Yields
The Improve step of the DMAIC process can be implemented on The PCBA Algorithm
in two different ways:
1. Through different weighting scenarios and general priority rules, over time the
algorithm can be Improved to optimize the priority list to ensure the most
important PCBAs are worked on first.
2. The firm's quality, yield, profit margins, and customer satisfaction metrics will
Improve by implementing the PCBA Prioritization Algorithm now and using the
results to fix the most important PCBA issues first.
4.4.1. Improvements using Different Weighting Scenarios
In section 4.3.3, the TOPSIS algorithm calculated weighted scores based on each
index having equal weightings. This section looks at different weighting scenarios to
determine the impact on the overall priority rankings. First, the PCBA Priority
Algorithm analyzed 20 random PCBAs as seen below in Table 32. The raw
manufacturing data was also randomized to protect Cisco confidentiality.
106
Table 32: 20 Randomized PCBA samples
PCBA Total Index Scoring System
Highest priority is placed on PCBAs with the lowest PCBA Total Index
PCBA Total Index Score [0 - 1001
Revenue and Demand Index Score 10 - 1001
PCBA #
PCBA #1
PCBA #2
PCBA #3
PCBA #4
PCBA #5
PCBA #6
PCBA #7
PCBA #8
PCBA #9
PCBA #10
PCBA #11
PCBA #12
PCBA #14
PCBA #13
PCBA #15
PCBA #16
PCBA #17
PCBA #18
PCBA #19
PCBA #20
It
Ouality Index Score [0 - 1001
I Mnsvnement
and Custnmer
I
. . . ~L·-·LII-·II- ~11~ VIYIVII--·L
Inder
Srnrp fi - 1i)01
-11~~~ UIVLI IV
-VVI
4
100
6
11
18
22
29
37
95
89
77
10
15
15
91 31
87 44
0
0
0
27
27
77 54 39
40 76 54 50
44 79 59 59
46 58 48 52
45 67 45 68
47
55
54
56
61
68
75
80
96
47
36
56
48
21
22
17
11
6
50
36
67
69
54
57
79
100
100
32
59
57
55
49
64
64
62
100
These PCBAs are prioritized from 1 - 20 based on equal weightings. Next, 17
different weighting scenarios were developed (see Table 33). These scenarios range
from dominant index weighting to a dominating factor weighting. Remember, the
factors make up the index weightings as defined by Equation 7. The equal index
weighting, labeled as Scenario I, is the base case.
107
Table 33: Different Weighting Scenarios
Scenario
I
II
III
IV
V
VI
VII
VIII
IX
X
XI
XII
XIII
XIV
XV
XVI
XVII
XVIII
Weightins for each
Index
(RDI, QI, CMI)
Description
Equal Weightings
RDI = QI
RDI = CMI
QI = CMI
Dominant RDI
Dominant QI
Dominant CMI
Dominant RDI Factor 1, a
Dominant RDI Factor 2, P
Dominant RDI Factor 3, y,
Dominant QI Factor 1, 6
Dominant QI Factor 2, F
Dominant QI Factor 3, C
Dominant QI Factor 4, sq
Dominant QI Factor 5, 0
Dominant CMI Factor 1, t
Dominant CMI Factor 2, K
Dominant CMI Factor 3, X
(33%, 33%, 33%)
(40%, 40%, 20%)
(40%, 20%, 40%)
(20%, 40%, 40%)
(90%, 5%, 5%)
(5%, 90%, 5%)
(5%, 5%, 90%)
(52%, 30%, 18%)
(52%, 30%, 18%)
(52%, 30%, 18%)
(18%, 64%, 18%)
(18%, 64%, 18%)
(18%, 64%, 18%)
(18%, 64%, 18%)
(18%, 64%, 18%)
(18%, 30%, 52%)
(18%, 30%, 52%)
(18%, 30%, 52%)
Weightings for each Index's factors
(a, p, y, 6, 1, 4, q, 0, I, K, -)
(11%, 11%, 11%, 7%,7%, 7%, 7%,7%, 11%, 11%, 11%)
(13%, 13%, 13%, 8%,8%,
8%, 8%, 8%,7%, 7%,
%,7%)
(13%, 13%, 13%,4%,4%,4%,4%,4%,
13%, 13%, 13%)
(7%, 7%, 7%, 8%, 8%, 8%, 8%, 8%, 13%, 13%, 13%)
(30%, 30%, 30%, 1%, 1%, 1%, 1%, 1%, 2%, 2%, 2%)
(2%, 2%, 2%, 18%, 18%, 18%, 18%, 18%, 2%, 2%, 2%)
(2%, 2%, 2%, 1%,1%, 1%, 1%, 1%, 30%, 30%, 30%)
(40%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%)
(6%, 40%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%)
(6%, 6%, 40%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%)
(6%, 6%, 6%, 40%, 6%, 6%, 6%, 6%, 6%, 6%, 6%)
(6%, 6%, 6%, 6%, 40%, 6%, 6%, 6%, 6%, 6%, 6%)
(6%, 6%, 6%, 6%, 6%, 40%, 6%, 6%, 6%, 6%, 6%)
(6%, 6%, 6%, 6%, 6%, 6%, 40%, 6%, 6%, 6%, 6%)
(6,6%,6%,6%, 6%,6%, 6%, 40%, 6%,6%,6%)
(6,6%,6%,6%, 6%,6%, 6%, 6%, 40%, 6%, 6%)
(6,6%,6%,6%, 6%,6%, 6%,6%, 6%, 40%, 6%)
(6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 6%, 40%)
Remember,
0, , K,1), and
The Total PCBA Index (TI) = f(a, P, y, 8, F,4, 0l,
1) the Revenue & Demand Index (RDI) = f(a, P,y)
2) the Quality Index (QI) = f(6, E,;, q, 0)
3) the Customer & Management Index (CMI) = f(t, K,,)
The PCBA Prioritization Algorithm then analyzed the 20 PCBAs from Table 32
across these 17 different weighting scenarios. Table 34 summarizes the findings.
Table 34: Different Weightings Scenarios Applied to 20 Randomly Sampled PCBAs
t'rtolttln esl II.BA acre diflerea Wegtln
Priorities
PCBA#1
PCBA#2
PCBA#3
PCBA#4
PCBA#5
PCBA#6
PCBA#7
PCBA#8
PCBA#9
PCBA#10
PCBA#II
PCBA#12
PCBA#13
PCBA#14
PCBA#15
PCBA#16
PCBA#17
PCBA#18
PCBA#19
PCBA#20
Shifts from I
Scn
rrioritizlng each PCBA across different Weighting .cenarles
I
(equal
weitting)
7
91
10
12
3
14
15
16
17
18
19
20
20
1
0
1
0
4
9
13
12
9
10
9
13
II
15
13
12
13
9
15
15
When compared to the original equal weighting base case in Scenario I, the shaded
areas indicate where the priority order remained unchanged with the new weightings.
The last row shows the number of shifts from the base case (Franklin & Niemeier,
108
1998). In each of the 18 cases above, the first, second, and 20th ranked PCBA
maintained their original order. This means that no matter what weighting the firm
decides to use, the priority algorithm will effectively prioritize the top 2 most
important PCBAs to work on while also placing the lowest priority PCBA in its
correct spot.
Fourteen of the eighteen cases maintain the top three PCBA order and thirteen of the
eighteen scenarios have the same bottom three PCBAs. Scenario V has the most
shifts, which is due to the dominant weighting on the Revenue & Demand category.
However, although this scenario has 13 shifts, the top 11 priority PCBAs are the
same, just in a different order. Scenario V merely shuffles around spots 3-9 and 12 17.
Additionally, the same weighting scenarios are analyzed across the 18 extreme comer
case scenarios 25 obtained from the DOE matrix. Table 35 shows the top priority
PCBAs are still ranked in the same spot across all the different weightings scenarios.
Granted there are more shifts than the randomly generated list, in all the different
weighting scenarios, the top 5 priority PCBAs are in the top 9 priority positions.
Table 35: Weightings Scenarios Applied to 18 Revenue Generating DOE Extreme Cases
Prioritizlng eachextremecase PCBA across differentWeightingScenarios
Priorities
PCBA#1
PCBA #2
PCBA#3
PCBA#4
PCBA#55
PCBA #6
PCBA #7
PCBA#
PCBA U9
PCBA #10
PCBA #11
PCBA #12
PCBA#13
PCBA #14
PCBA#15
PCBA#16
PCBA #17
PCBA #18
Shifts
froml
(equal
weightings)
1
2
3
4
5
6
7
8
9
10
II
12
13
14
15
16
17
I8
0
II
I
4
2
6
5
10
3
12
8
11
7
16
14
9
13
17
15
11
1
IV
3
3
4
2
7
8
9
5
6
13
14
10
11
12
17
15
16
4
6
5
3
10
8
12
7
11
9
14
16
13
15
V
1
2
3
10
4
5
6
II
12
7
8
13
14
15
9
16
7
18
17
18
18
12
15
11
16
12
VI
I
7
2
3
8
13
5
9
4
14
10
16
11
6
15
17
1
12
VII
15
11
1
2
7
3
8
5
13
4
9
10
14
6
11
16
15
12
VIII
I
3
2
10
4
7
5
13
11
8
6
16
14
12
9
17
15
IX
X
XI
3
2
10
4
7
5
13
II I
8
6
16
14
12
9
17
15
3
2
10
4
7
5
13
11
8
6
16
14
12
9
17
15
7
2
3
8
13
5
9
4
14
10
16
11
6
16
16
16
1
17
12
II
6
15
17
12
XIII
I
7
2
3
8
13
5
9
4
14
10
16
11
6
15
17
12
XIV
I
7
2
3
8
13
5
9
4
14
10
16
11
6
IS
17
12
15
15
15
15
15
XII
II
7
2
3
8
13
5
9
4
14
10
XV
1
1
7
2
3
8
13
5
9
4
14
10
16
11
6
s
17
12
XVI
I
3
7
2
9
5
13
4
8
11
15
6
10
14
16
12
17
XVII
I
3
7
2
9
5
13
4
8
I1
15
6
10
14
16
12
17
1
14
I4
15
14
14
XVIII
I
3
7
2
9
5
13
4
8
11
15
6
10
14
16
12
17
is
14
25 18 corner cases are the revenue generating scenarios from the 27 possible corner cases. The non revenue
generating cases will get prioritized into the "No Demand" category.
109
Overall, the weighting schemes can be changed to improve the overall priority list
depending on the relative importance of each factor. However, weighting the factors
has the most impact on the PCBAs that are ranked in the middle of the list. The very
lowest scores remain low while the very highest scores will remain high. Therefore,
to improve the PCBA Prioritization Algorithm further, the firm should pilot the
program to determine if, over time, the PCBAs ranked highest are in fact the right
ones to be improving. Additionally, overall PCBA yields should improve as costs
decline. These critical metrics should be monitored and reviewed through the
process. Thus, trying to Improve the weightings ties right into the Control part of
DMAIC (see section 4.5), because one will need to monitor the PCBA over time to
determine, if, in fact, its index scores and subsequent yield and cost metrics are
improving.
4.4.2. Improve PCBA Yields Today
Using the PCBA Prioritization Algorithm, the firm can start to see results today.
Working on the most important PCBAs now will allow improved yield gains, more
revenue, and improved customer satisfaction to be realized sooner. Additionally,
time will show the true return on investment of the PCBA Priority Algorithm. The
current situation has no global method to prioritize PCBAs. Because this algorithm
offers the first unified prioritization approach, any method is bound to reap some
benefits over no method. A pilot is now underway at Cisco, and a portion of the
PCBA Prioritization Algorithm is in place today via a web based application.
4.5.
Control: Monitor PCBA Metrics over Time
The algorithm should be used as a tool to effectively allocate manufacturing resources to
the highest priority PCBAs. The algorithm allows the entire organization to globally
prioritize while also enabling different manufacturing sites to prioritize the same way
locally. If each site prioritizes the same, the manufacturing firm will effectively be
working on all the PCBAs with the highest priority. The algorithm should be able to run
in real time; however, the priorities should be set on certain time intervals that allow
110
previous issues to be resolved. Therefore, the time interval for different companies may
be different. Based on the typical cycle time to complete a yield improvement plan, the
re-prioritizing of the algorithm can be determined.
Over time, the Total PCBA Index score, along with the Revenue & Demand, Quality, and
Customer & Management Index scores can be tracked for each PCBA in the portfolio. If
the algorithm works effectively, the number of low scoring PCBAs should decrease as
more and more improvements are being made. The goal should be that all PCBAs in the
portfolio achieve a 60 process. If so, every PCBA with a positive demand profile would
receive a score of 100. The firm should track the index scores by PCBA and set
appropriate goals. Using the PCBA Priority Algorithm gives the manufacturing
organization a better handle on how to improve PCBA yields and drive costs lower.
Understanding these knobs will enable the firm to remain competitive and take the proper
next steps to achieve a 60 yield process on every single PCBA manufactured.
4.6.
Summary
The goal of the PCBA Prioritization Algorithm is to provide a tool for the firm to
improve overall PCBA yields. With thousands of PCBAs being manufactured,
determining which PCBAs to improve first can be quite cumbersome and very difficult.
Based on current best practices, this approach is a first step to standardize the
prioritization methodology across the company. The method looks at manufacturing
factors that determine a PCBA's overall revenue generation potential to ensure the firm is
leaving no profits on the table. A PCBA's quality level is measured to sort out which
have the most potential room for improvement. Finally, how the customer and
management perceive the PCBA's performance can affect where it ranks in the priority
list. Using the Six Sigma DMAIC approach, the PCBA Algorithm calculates indices that
give employees across the firm the ability to quickly and easily determine which PCBAs
to work on that will have the greatest positive impact on the overall business.
111
This page has been intentionally left blank
112
5. Chapter 5: Organizational Design and Implementing Change
It is imperative to understand the nuances within any organization when driving change.
The majority of this thesis was spent working on a significant change initiative at Cisco.
Therefore, before implementing any ideas developed in this thesis, knowing how the
organization will react to the change becomes extremely important. To become a world
class organization and achieve a 6a manufacturing process, significant process changes
are required; but, more importantly be successful, the entire company needs to be
onboard with the idea of change. Chapters 3 and 4 discussed how certain tools like the
Component Quality Yield Optimization Model and the PCBA PrioritizationAlgorithm
will help in the journey to achieve a 6a yield process. However, how will the
organization react to implementing these ideas and will the company change and adopt
the tools as part of their new mode of operation? Let's look closer at Cisco's
organizational design and discuss how it will react to change.
5.1.
Company Intro
Cisco's Vision is to change the way we live, work, and play.26 Cisco's Mission is to
shape the future of the internet by creating unprecedented value and opportunity for their
customers, employees, investors and ecosystem partners. 27 Today, the end user connected
to the internet is empowered to drastically change how they communicate with their
social networks, creating new forms of communication through collaboration no matter
where they are located in the world. A social network is a social structure made of nodes
which are generally individuals or organizations defining the manner in which we stay
connected through specific types of interdependencies like family, friends, communities,
institutions, commerce, and interests.28 Thus, through Cisco, the social network has
evolved to the human network.
26 Chambers, John. Live Presentation. CEO of Cisco Systems, Inc. 2007
27 Chambers, John. Live Presentation. CEO of Cisco Systems, Inc. 2007
28 http://en.wikipedia.org/wiki/Social_network
113
Like a social network, the human network connects people but differently than before.
The human network relies on the internet network which enables a new way for humans
to connect, create, collaborate, and communicate. Cisco believes this network is the
platform enabling all forms of collaboration allowing a seamless convergence of voice,
video, data, and mobility to create an entirely new life experience.
Cisco designs, manufactures, and markets internet protocol (IP) based networking and
communications solutions such as routers, switches, and phones. Cisco is a very large,
global company with over 50,000 employees, of which 9,000 are in manufacturing. As
of 2007, Cisco has an $186B market cap with about $40B in revenue. Since 1993, Cisco
has made 125 acquisitions that make up 30 different business units.
Design is done in-house; however, 100% of the manufacturing is outsourced to contract
manufacturers (CM) around the globe. Additionally, there is a vast array of supply chain
complexity. Cisco deals with a quick lead time and a configure-to-order supply chain.
The four customer segments (commercial, service provider, enterprise, and consumer)
deal with a wide range of product complexity. Each quarter over, 250,000 orders are
processed. There are 196 active product families with 23,000 product identification
numbers and 600 suppliers with 50,000 purchased part numbers. 29
Cisco has a significant world wide footprint. Including the recent acquisitions of
Scientific Atlantic and Linksys, there are 41 sites, 4 CMs, 13 original design
manufacturers, and 19 logistic centers. 30 Cisco segments its products into low-, mid- and
high-end. The processes are tailored for these different segments and either operates as
make-to-stock or make-to-order.
In the following sections, I will analyze Cisco using MIT Sloan's Three Lens Analysis.
This analysis is based on John Carroll's work entitled "Introduction to Organizational
Cisco Systems, Inc. Internal Presentation. 2007
30 Cisco Systems, Inc. Internal Presentation.
2007
29
114
Analysis: The Three Lenses" and the teachings from Professor Kate Kellogg in the
Organization Processes course at MIT Sloan. The Three Lens analysis looks at an
organization from the strategic, political, and cultural perspective and is cognizant that
each lens provides insight into how and why an organization functions the way it does.
The Three Lens Analysis is a framework that does not necessarily provide answers to
solving complex problems, but does provide insight into the implication of change within
an organization.
5.2.
The Strategic Lens
The strategic lens views the organization as a machine, focusing on how a company
rationally organizes, links, and gives the proper incentives to its employees to accomplish
the end goal demanded by the market. John Carroll describes the strategic lens in further
detail:
"People who take this perspective view the organization as a kind of machine that
has been designed to achieve goals by carrying out tasks. The designers of the
organization, the Board of Directors and senior managers, have a strategy or
purpose for the organization based on rational analysis of opportunities and
capabilities." (Carroll, 2006, p.3)
Additionally, Professor Kellogg suggests the strategic lens is very attractive to people
with a technical background, thinking "if only the gears would mesh together properly,
the company would be successful.""' Professor Kellogg suggests analyzing the strategic
lens with the following framework: (1) Organizations are machines, (2) Mechanical
systems are crafted to achieve defined goals, (3) Parts must fit well together and match
environmental demands - the organization must be grouped, linked, and aligned to the
goals, and (4) Action comes through planning. 32
Cisco aspires to maintain its leadership position as a global networking company. Cisco
intends to transform life's experiences by making the network the platform - seamlessly
3' Kellogg, Kate. Organizational Processes Lecture Notes. Fall 2006
32 Kellogg, Kate. Organizational Processes Lecture Notes.
Fall 2006
115
combining voice, video, data, and mobility. The CEO of Cisco, John Chambers, believes
"Cisco's strategy is a story based on change. Through multiple transitions in the last
decade and over the next 3-5 years, the network will evolve from the plumbing of the
Internet- providing connectivity--to the platform that enables people to experience
life." 33
Like a machine created to design and manufacture innovative business solutions, Cisco
operates by grouping its people accordingly to sustain its competitive advantage. At the
executive level, the firm is specifically split into six main organizations (1) Cisco Design
Organization; (2) Operations, Processes, & Systems; (3) Marketing; (4) Customer
Advocacy & Global Center Operations; (5)World Wide Sales; and (6) Finance, each run
by an Executive Vice President. 34 This functional organization will design and deliver
new products driven by the technology and business environment across the globe to gain
market share while growing profits.
Cisco's Design Organization (CDO) is organized functionally by business unit (BU),
each led by a vice president. Each BU may contain one or more product families and
each product family may contact one or more products. Manufacturing is also organized
functionally by department and each led by a vice president. Each department for the
most part, is grouped by focus areas such as test development, yield, and quality control
to name a few. Certain groups and skills are centralized in both CDO and manufacturing,
but it is unclear how they link to each other. This ambiguity causes inefficiencies within
the company, such as duplicating the same work, not communicating well between
groups, prioritizing goals differently across the company, setting wrong incentives, and
identifying incorrect problem owners.
The paradox is that Cisco is a networking company designing and manufacturing
networking gear to enhance world-wide communication, but employees do not network
33Chambers, John. CEO of Cisco Systems, Inc. Internal Presentation. 2007
34 In 2007, the Executive Vice President (EVP) became a new role at Cisco. This will be discussed further
in the cultural section of the paper.
116
as well as they could across functional groups. On the whole, Cisco divisions, business
units, organizations, and departments remain very segregated. The Cisco management
team recognized this problem and is currently working to break down these
communication barriers and organizational silos. Integration teams began forming
recently suggesting that the company may be moving from a functional organization to
more of a matrix structure. These new cross functional teams appear to be working and
may help solve the functional silo problems in the short term, but ultimately, Cisco must
keep supporting a cultural shift from segregated silos to a more collaborative culture
across the company.
My internship, entitled "Test Excellence: Modernizing the Test Strategy and Governance
Model to Enable an Agile, Aligned, and Adaptive Supply Chain," focused on designing a
world-class test strategy at Cisco to guarantee sustained innovation and competitive
advantage for the next 5-10 years. My internship spanned across two main divisions, one
main organization, and three main departments, as shown below with respective leaders:
* Division 1: Cisco Design Organization (CDO), Executive Vice President
* Division 2: Operations, Process, & Systems, Executive Vice President
a. Organization 1: Manufacturing, Senior Vice President
i. Department 1: Technology & Quality, Vice President
ii. Department 2: Product Operations, Vice President
iii. Department 3: Manufacturing Operations, Vice President
The BUs are responsible for designing new products that meet the needs demanded by
the customer. The BUs work with the customer advocacy and marketing groups to
develop new products. There are pockets of central groups within the CDO to help share
best practices, but these central teams do not operate as efficiently as they could and do
not talk to every BU.
Corporate Quality; Customer Service; Information Technology (including the Chief
Information Officer); Strategic Planning; Human Resources; Corporate Affairs and Legal
Services & General Counsel all report to the Operations, Processes, & Systems Division.
A Senior Vice President runs Manufacturing, who also reports to the Operations,
117
Processes, & Systems division. Beneath the manufacturing organization, ten Vice
Presidents are respectively responsible for Demand Management & Planning; World
Wide Supply Chain Management & Advanced Sourcing; Global Business Operations;
Mergers & Acquisitions; World Wide Reverse Logistics; Global Commodity
Management, Technology & Quality; High End Product Operations; Low End Product
Operations; and Manufacturing Operations.
People are grouped by skills within each of these departments under the manufacturing
organization. For example, Table 36 below displays how test engineers are found in four
different departments in the manufacturing organization and have very different
responsibilities.
Table 36: Test Engineer Responsibilities within Different Manufacturing Departments
Manufacturing
Test Engineer Responsibilities
Department
Product Operations
Plan, test, design, develop,
procure equipment, own product
Manufacturing
Sustain, enhance, capacity
Operations
Technology & Quality
Develop process, strategy, and
execution
Contract Manufacturers
Sustain & troubleshoot
Located all over the world, these test engineers do not necessarily communicate or share
best practice efficiently. On one hand, Cisco has no good linking system or forum for the
test engineers to discuss issues. Products with similar architecture and components may
use completely different tests in manufacturing, while different products may get a
cookie cutter test plan from a completely different product line. Additionally, test scripts
are written differently for different products at various contract manufacturers. Each
group completes its functional role differently in this silo'd world, causing inefficiency in
the entire system. On the other hand, there are other manufacturing forums that do share
118
best practices and leverage existing data effectively. Cisco continues to identify these
best organizational practices and change the organizational structure to improve and
better utilize its employees.
Overall, Cisco employees are beginning to understand how Cisco's mechanical gears
really mesh together. The goals and initiatives are evident on the badges the employees
wear. Also, the CEO and Manufacturing Senior Vice President began a Manufacturing
Excellence initiative three years ago to align the organization to enable Cisco to
capitalize on the anticipated growth of Web 2.0.35 ,36 In the last three years, quality and
cost have become a major concern for Cisco but alignment to these goals has lagged.
The old incentive structure rewarded time to market goals. This structure promoted many
people to their current day vice president and executive positions but does not work well
in today's quality focused manufacturing environment.
In an effort to support the overall Manufacturing Excellence initiative, Cisco's executive
team changed the incentive structure last year to better align the employee priorities to
overall company goals. Collaboration, teamwork, and quality are a main focus for each
employee's performance review. Quality, although very important, never was the #1
priority for the design group - it had always been time to market. This year, the CEO is
working to change the incentive structure to reward the design group based on the quality
of the product they deliver rather than the speed at which they deliver. Additionally, the
CEO is widely respected and trusted. He cares about Cisco employees and collaborates
with employees through on-site presentations, video blogs, and Cisco's online TV
35 According to wikipedia, "Web 2.0, refers to a perceived second generation of web-based communities
and hosted services - such as social-networking sites, wikis, and folksonomies - which aim to facilitate
collaboration and sharing between users." http://en.wikipedia.org/wiki/Web 2.0
36 According to a 2007 Cisco presentation, "Web 1.0" is "the original plumbing of the internet: the pipes,
connectivity, personal computers, and transporting of data. Web 1.0 provided the pipes to connect people
with personal computers to the World Wide Web-transporting data around the globe and enabling
pervasive and ubiquitous e-mail, e-commerce, instant messaging and other Web-based applications."
119
network, called Cisco IPTV, ensuring everyone is aligned to a "one team, one goal"
philosophy.
Product lifecycle activities are mainly linked through product operations; however, when
asked who owns a product, everyone says they do. During new product development,
CDO engineers work directly with the central test team under Manufacturing's
Technology & Quality department to develop the test plan for each new product. When a
new product is ready to be released to market, CDO works with new product introduction
engineers (NPIEs) under product operations. After a successful new product
introduction, the product is handed over to manufacturing operations to sustain the
product through its product lifecycle. At this point, the NPIEs, who are the major link
from design to manufacturing, often drop out of the communications loop to work on the
next new product. When this occurs, there is no direct owner, which creates ambiguity in
the system and leads to no official ownership because everyone thinks the other person
will take responsibility. As a current pilot suggests, when NPIEs take ownership of the
product from the beginning of life to the end, the product's metrics are better.
Cisco defines the roles and responsibilities through a RACI37 document, but people often
change roles within a department or organization. All documents and information are
online, but employees often question who reads, understands, and abides by the
documents. For instance, when asked who specifically owns quality or yield, no one
really knows. There is a quality group in Manufacturing Operations, but also a quality
and yield group in the Technology and Quality. Cisco recognized this issue and has
started to better align people's roles and tasks, specifically ownership of problems in and
around quality issues. Through the Test Excellence change initiative, the new focus on
quality and yield improvements provides a terrific venue to adopt the tools of this thesis.
37 RACI is Cisco's roles & responsibilities matrix.
(R = Responsible IA = Accountable IC = Consultant I
I = Inform.)
120
Additionally, in San Jose, CA alone, Cisco employees are spread across 40 buildings
while over 20 major satellite facilities are located around the globe. It is inconsistent how
people are grouped at each building: some groups reside on the same floor, while some
groups are spread across an assortment of buildings on and off campus. Also, many
people telecommute. This physical layout inhibits how work gets accomplished and also
contributes to the silo'd culture apparent at Cisco.
Cisco has significantly grown and captured enormous value with the technology the
company brings to market. To successfully fit into the environmental demands, Cisco
planned and defined its business model. The strategy was to find great intellectual
property, acquire it, integrate the innovation into the overall business model, and then
outsource the manufacturing. Cisco has done a fantastic job over the years, but the
acquisitions have created a very silo'd workforce, especially within CDO. On one hand,
some BUs display a "not-invented-here" attitude which creates difficulties for the central
groups to make any changes. On the other hand, very successful acquisitions have
created a "my-way-or-the-highway" attitude. This interesting dichotomy of social
networks within the company has molded the political and cultural landscape at Cisco.
These strategic lens issues shape a part of Cisco's culture and will be discussed in further
detail in the cultural lens section.
The Test Excellence initiative is one of the first highly collaborative Cisco project to span
the major design and manufacturing divisions: CDO, product operations, technology &
quality, and manufacturing operations. The current test strategy was intended to get
Cisco to $40B in revenue by 2004. The current 223 page strategy document, like the
history of Cisco acquisitions, did its job and was a success - but it just will not suffice,
especially with the anticipated growth driven by Web2.0. The "test-the-heck-out-of-it"
strategy is considered "ad-hoc" and is just old. Cisco needs to "Adapt or die" (Ashmore,
2000, p. 25). Cisco is evolving with the times in order to fit profitably into its future
competitive environment.
121
In summary, "orgs are fundamentally rational, in the sense they can be designed to
achieve shared goals" (Carroll, 2006, p.5). The strategic lens identifies Cisco's
functional structure is designed to support the company's main objective: To design and
manufacture products at 6a quality levels to support the network of people connected to
the internet. However, the functional structure lacks proper linking mechanisms and
incentives to align how its employees efficiently work. Despite the complexities within
communication channels across the organization and the many silos that have been
formed, Cisco's business has been very successful thus far. In order to remain
competitive, Cisco's recent goal is to collaborate better across functional groups. The
Test Excellence Initiative is doing just that and is a collaborative team allowing
employees to effectively work across groups and share best practices. The strategic lens
suggests that Cisco needs to better align the groups and give proper incentives to
collaborate to ensure the Test Excellence change is more of a long term solution rather
than a short term fix. The Component Quality Yield Optimization Model and PCBA
PriorityAlgorithm, which were born out of the Test Excellence Initiative at Cisco, are
initial steps to collaborating across groups and sharing best practices. With the proper
incentive structure to achieve a 6a yield process for all products, from a strategic lens
perspective, the firm should react positively to change and adopting the two analytical
tools discussed in this thesis. Let's see how the political lens may be affected.
5.3.
The PoliticalLens
The political lens looks at the political and bureaucratic nature of an organization. John
Carroll describes the political lens:
"The political lens shatters the assumption that an organization has goals that 'it'
is pursuing. Instead, people who use the political lens view the organization as a
contested struggle for power among stakeholders with different goals and
underlying interests. Whereas the strategic design lens groups and links units that
must work together to accomplish tasks, the political lens combines units with
similar interest and goals into coalitions that advocate their side of important
issues. Goals and strategies are either imposed by a ruling coalition or negotiated
122
among interest groups. At circumstances, power shifts and flows, coalitions
evolve, and agreements are renegotiated." (Carroll, 2006, p.5)
Additionally, Professor Kellogg suggests "the [political] lens is about contests and
conflicts" and can be analyzed with the following framework: (1) Organizations are
contests, (2) Social systems encompass contradictory interests, (3) Competition for
resources is expected, and (4) Action comes through power - and there are various
sources of power.38
Cisco's political spectrum ranges across the company, but the power definitely resides at
the headquarters in San Jose, CA. With over 50,000 employees across the globe,
coalitions and networks have emerged since day one. The biggest source of power still
involves the employees from the most successful acquisitions. According to outsiders on
the inside, acquired employees that helped launch Cisco into the powerhouse
organization it is today retain the most power and are part of the esteemed "boys clubs"
and "circles of trust." Those people are known to stick together. However, as the market
has evolved since Cisco's incorporation in 1984, their legitimacy appears to be expiring.
Power has slowly started to shift from these once powerful acquired employees. While
these old-timers may have laid the groundwork for Cisco's current success, Cisco needs
to change to remain successful in the years to come.
Cisco has three main sources for power. The first source of power is role-based. Any
employee looking through Cisco's online active directory can see the reporting structure
of anyone in the company all the way to the CEO. From an initial glance, the closer to
the CEO, the more responsibilities the employee possesses, and thus, the more power,
expertise, and recognition that employee assumes. Whether the power is role-based or
structural, the amount of power an employee holds waterfalls down through the ranks.
Employees aspire to become directors, vice presidents or even higher. They know when
current holes in the organization will open and when current leaders are scheduled to
retire or leave.
38
Kellogg, Kate. Organizational Processes Lecture Notes. Fall 2006
123
The second source of power is strong personal networks. Many personal networks were
established within the acquired company before integrating into Cisco. In many cases,
after the acquired company splits up in the organization, the power of the informal ties
within the split group grows stronger. The power does not tie directly to the status the
online active directory suggests but correlates more to the informal personal ties. Since
the personal ties can be stronger than the ranks, employees often circumvent the direct
reporting structure to get tasks accomplished quicker. Or, because of some close
personal networks, employees may give the "inside scoop" from their cronies at the top.
However, some personal networks were also established in-situ within Cisco. There is
fierce competition for human capital and through these personal networks, directors and
VPs came to know good and bad employees, cherry-picking people to lead or vetoing
current leaders on high priority projects. Similarly, managers strategically place "their"
people in specific meetings to act as a voice for the group in order to keep the perceived
power at a balance.
Position is power at Cisco, but so is data. Data is the third source of power. In the very
early days, people were promoted to VPs and to higher ranking manager positions for
making the right decision to ship the product, even at the expense of quality. Often
times, the product was not ready for the market, but the decision remained to ship the
product due to customer demand. These decisions were data driven by the market
demand, and quality issues often inhibited the product performance in the field. As more
and more products were shipped to customers because of a Vice President's decision, the
more the revenue and the VP's power grew. The stock soared; no one could lose. Since
Cisco's initial public offering in 1990, the stock, along with employees' egos, surged
75,000% until the internet bubble burst in 2000 (O'Neil, 2004, p. 35). After the stock
decline, poor quality started to become an issue. Today, quality remains a major concern,
and employees are not promoted for just shipping products as in the past. The ongoing
political game continues - one that now involves new egos battling it out on how to
124
change an engrained "time-to-market" culture to one that delivers "world class, Six
Sigma quality."
Within Cisco, CDO historically has carried the most power and still does today. There is
an apparent contest between CDO and the rest of Cisco's divisions. CDO believes they
are the innovation muscle and the mastermind behind Cisco's growth, revenue stream,
and high margins; however, since manufacturing makes the product, they are responsible
for the success since they deliver the volume. Both groups are right but do not easily see
the success from each other's perspective. CDO is driven by time to market goals while
manufacturing is driven by cost, quality, and delivery goals. These goals inherently
compete with each other and are fundamental flaws igniting the continuous cultural clash
between the two divisions. The organizational structure just prevented the key
stakeholders from really understanding each other's goals, which ultimately caused more
finger pointing, a greater internal power struggle, and less agreeing.
Cisco is a networking company, so although people do not necessarily communicate well
across functional groups in the strategic lens, social networks have formed and will
continue to grow. Social networks are different than the functional networks. From a
strategic lens, it may seem like the company struggles to accomplish its main objectives.
However, after viewing the company through a political lens, it's easier to see how the
critical tasks get done. The political and social networks are an internal web, linking
employees across the company. Employees will use the political landscape to
accomplish tasks they deem important. CDO maintains both the "not-invented-here" and
"my-way-or-the-highway" political agendas while several employees' egos remain
elevated and overpowering. Slowly, Cisco as a whole, is working to break down these
attitude silos and move to thinking and working for a one vision, one goal philosophy.
As the company remains more quality focused with certain managers pushing quality
politics through the Test Excellence Initiative, the Component Quality Yield Optimization
Model and PCBA PrioritizationAlgorithm should be accepted over the long run only if
the Test Excellence initiative remains successful, senior management continues to
125
support it, and the rest of the company embraces the change. "In summary, the political
lens assumes that any organization is a diverse collection of stakeholders with different
and sometimes conflicting interests. The organization is heavily influence by those with
power, ruling coalition, but power is constantly shifting and being contested. Not only do
different groups hold different amounts of and kinds of power emanating from different
sources, but as the environment shifts or new strategies are developed, groups come to
the fore that have the capabilities to deal with these new demands" (Carroll, 2006, p.7).
5.4.
The Cultural Lens
MIT Professor Shoji Shiba teaches that one can learn a great deal about a company's
culture by looking at the visible, invisible, and unknown. Observing a company at work
or on the factory floor during day-to-day tasks, meetings, planning sessions, etc, allows a
unique perspective into how the company's culture influences how works gets
accomplished. John Carroll further describes the cultural lens:
"Those who use the cultural perspective assume that people take actions as
functions of the meanings they assign to situations. We all make sense of
situations in terms of our past history, analogies and metaphors, language
categories, our observations of others, and so forth. These meanings are not given,
but rather are constructed from the bits and pieces of social life."(Carroll, 2006,
p.8)
As Professor Kellogg teaches in her class, one can see the visible, invisible, and unknown
by identifying the company's espoused values and traditions, artifacts, and basic
assumptions - both conscious and sub-conscious: "the [cultural] lens is hard to see" and
can be analyzed using the following framework: (1) Organizations are institutions, (2)
Symbolic systems are meanings, artifacts, values, and routines, (3) Informal norms and
traditions exert a strong influence on behavior, and (4) Action comes through habit. 39
Cisco is an institution that is growing up. Employees compare the evolution of Cisco to
that of a human. In its younger days, Cisco could do no wrong. The child-like, immature
39 Kellogg, Kate. Organizational Processes Class Notes. Fall 2006.
126
company was free-spirited and composed of entrepreneurial and innovative people. The
firm spent money at will and grew rapidly. At one point, Cisco hit an enormous growth
spurt and hired 1,000 people every month for 12 months in a row. 40 Still young, the
organization developed habits such as not prioritizing quality or cost in order to succeed.
These habits were appropriate for Cisco, the start-up. The Cisco culture became a residue
of its success. The internet bubble burst, and although Cisco felt the pain of the deflation,
the teenaged company endured the hardships that followed.
Now 20 years old, Cisco is at the similar college graduate life stage, trying to figure what
to do when it enters the "real world." The college graduate faces choices such as finding
the best job in the best location with the end goal, among other things, to earn a living,
add value to society, and make his/her parents proud. Today, Cisco faces similar choices
with the goal to capitalize on new products for its graduation into the "real world" of
Web 2.0 and beyond.
At Cisco, appearance counts. Thus, Cisco employees on the whole tend to dress a step
higher than business casual. The appearance transcends clothing attire and can be seen
from clean and sleek PowerPoint presentations to visually stimulating customer
experience centers to the continuous building renovations. The new Cisco logo is used
everywhere with pride. Presentations are animated and colorful. One employee
emphatically documented that "Everything's in PowerPoint. If it is not in the deck or it is
not presented, the data does not exist. By the way, it also helps to have nice looking slides
' 41
too."
John Chambers, the CEO of Cisco, epitomizes Cisco's culture. He is a true Cisco
employee, practicing what he preaches by collaborating with employees and leading with
passion and vigor. He speaks the Cisco languages and lives the Cisco way of life. He is
a great salesman, intelligent, well dressed, and well spoken. Chambers is a true leader
40 Interview with Employee. 2007
41 Interview with Employee. 2007
127
who puts the customer first: "Customer success and satisfaction are at the heart of Cisco's
business strategy and key drivers of our current and future success. There is over a 10
year history of formally tracking customer satisfaction, which is tied directly to the
employee bonus plan." 42
Thirteen espoused values build the foundation of Cisco's culture: Innovation, Continuous
Improvement/Stretch Goals, Quality Team, No Technology Religion, Profit Contribution
(Frugality), Giving Back/Trust/Fair/Integrity, Teamwork, Market Transitions, Fun, Drive
Change, Empowerment, Open Communication, and Customer Success. Each Cisco
employee wears these values on a badge.
One important espoused value missing is the need to be data driven. To date, Cisco has
neither been a technocratic or experience-based firm (Go, 2007). On one had, "[f]irms
that are founded by technologist typically place a high value on engineering or analytical
thinking. To be heard, one must first build credibility through a proven track record of
technical competence" (Klein, 2004, p.77). On the other hand, experience-based firms
"tend to use seniority, age, and company longevity as a basis for valuing an employee's
worth and knowledge" (Klein, 2004, p.78). Because Cisco has acquired 125 companies
in the last 20 years, the company has also acquired 125 different cultures, not to mention
125 different decision making processes. Some acquisitions have integrated rather well
and some have not. In fact, the continuous acquisition culture has formed many of
Cisco's traditions, values, and artifacts.
Although data is not an espoused value, Cisco is a culture of data junkies, over analyzing
the terabytes of data collected. According to one employee, Cisco "has very low
organizational memory." The same data analysis may be completed time and time again
by either the same or different employees. Then, the data is presented repetitively at the
same or different review meetings. Data tools and web dashboards have been created and
re-created to alleviate the over-analysis, but never standardized. Additionally, with all
the terabytes of data collected, employees still have trouble drawing conclusive actions
42 Chambers, John. CEO of Cisco Systems, Inc. Internal Presentation.
2007
128
from the data. Hence, on a high level Cisco makes decisions neither via data nor
seniority. Instead, the company has grown to use both decision making processes and
remains somewhere in the middle of the decision making spectrum - a collaborative
mash-up of data and seniority.
Company artifacts decorate the buildings. Foliage, shrubs, and waterfalls landscape the
buildings and outside walk ways nicely. The CEO's building has the biggest waterfall
and best landscaping by far. Running paths with exercise stations twirl through the
campus. Several gyms and classes are available for employees to use. The Cisco colors
decorate the buildings, each looking the same. Each building is built with essentially the
same layout. A majority of the buildings, mostly for CDO groups, are on a street named
Cisco Way. Walls are plain and simple, dressed with nice art pictures and the internally
famous great engineering methodology (GEM) diagram.
GEM is a Cisco tradition - it's a fundamental planning tool to how things get done.
Every Cisco employee knows, understands, and follows the GEM process. The GEM
process is a basic product development plan, but the milestones can be used for anything
whether designing a new product to managing a program. Each GEM milestone is
known and is a basic assumption that every employee uses it when planning, doing, and
reporting their work. The GEM process allows any Cisco employee to know exactly
where a plan is at and standardizes how work gets done across the many different Cisco
silos.
The meaning of the Cisco name is tradition. Cisco was derived from the name of the city
San Francisco. The logo for the company has been based on the Golden Gate Bridge and
really encompasses the idea that Cisco is a networking company bridging people all over
the world together. Traditional spellings used to start Cisco with lower case "c" to
remind employees of the company's roots. 43 As Cisco tradition continuously innovates,
the logo also changes to represent the new Cisco. Recently, Cisco changed its logo to
43 Cisco welcome document.
129
represent a gateway to the future and has deemed its new phase as "Cisco 3.0." The logo
is everywhere and the new image still resembles a bridge, but also looks like blinking
eyes symbolizing Cisco's dedication to bridge voice, video, data and mobility around the
globe while providing the essential bridge to the future. Cisco understands that today's
global end user is empowered to significantly impact our lives by connecting and
collaborating online. The new, refined Cisco will stay one step ahead of current Web 2.0
technologies. In fact, Cisco's culture is composed of Web 2.0 technologies, as the CEO
promotes online collaboration, using the world as a lab to innovate and grow, creating
more end-to-end product solutions, and innovating on the convergence of voice, video,
data, and mobility. Cisco even has a "Change Management Team" to ensure the proper
amount of change is occurring within the company. In fact, change is the only constant at
Cisco. However, while Cisco is trying to connect the world, its culture still remains very
structurally, functionally, and virtually silo'd.
Structurally, over 40 buildings in San Jose as well as numerous buildings around the
globe physically separate people, causing less and less face time interactions. It appears
that every time Cisco acquired a new company, a new building was built. Then, a new
business unit was developed and added to the CDO organization structure. Acquisitions
appeared to be somewhat "ad-hoc" in the sense that Cisco realizes some acquisitions
would be successful and some would not. The successful acquisition contributed to the
successful social and political networks and coalitions that formed within Cisco. Overall
acquisitions were very successful for Cisco's revenue stream.which contributed to the
power the CDO gained. CDO has been king at Cisco. In fact, CDO is the only
department within Cisco that uses the Cisco name: Cisco Design Organization. There is
no Cisco Manufacturing, just manufacturing. There is no Cisco Product Introduction,
just product operations. The CEO is trying to change the culture. Now, the Cisco made
of many different BUs, divisions, and departments is trying to become "One Cisco."
Teamwork is now an espoused value written on all badges and at the top of every
employees performance review. To be successful, every employee needs to work for one
Cisco, one vision.
130
A majority of the buildings have a manufacturing floor or lab on the first level. However,
since Cisco out-sources 100% of its manufacturing to CMs, most manufacturing floors
are not used, and have an eerie ghost town feel. Products that used to be shown to the
customer are now only displayed in certain customer relation buildings. These
experience centers are the most exquisite and visually stimulating places on campus.
In each building, 30 - 40 enclosed offices are in the middle of each floor in a 2x3 layout.
Managers, directors, VPs and senior employees occupy these offices. The offices have
shades that can be shut creating an initial sense of a very close-door culture; however,
Cisco is very open-door and welcomes impromptu meetings. The rest of the employees
sit in cubicles that occupy the remaining floor space. This creates a natural divide
amongst the ranks.
The structural barriers have given rise to many similar habits among Cisco employees
that have added to its culture. It is common for people to call into meetings across
campus rather than take the 15-20 minutes to commute to the specified meeting place.
Cisco is a culture of multi-tasking. Employees may call-in to start the meeting on time
and show up physically 30 minutes in the same meeting. On the San Jose campus,
employees drive between buildings all day. Usually, employees will walk to nearby
buildings, but since most are running late, a quick drive will save time. To communicate
with Cisco employees, one must email, instant message (IM), call, and text the person
because it is nearly impossible to know where that person will be. A typical sequence
may be something like this: start by IMing, then calling and/or texting. Finally, email to
document the information transaction.
Functionally, Cisco is very silo'd amongst its division, business units, and contract
manufacturers. Although various central organizations exist, little communication
actually occurs across the company. Integration teams are newly being formed to help
drive best practices in each silo and socialize the best practices amongst Cisco. However,
caused by the Cisco culture of "it wasn't invented here" or "my-way-or-the-highway"
attitudes discussed in the Political Lens, the socializing of best practices is met with
131
resistance and implementation is proving to be difficult. Hence, following through with
the last 20% may be more difficult due to political pressure than cultural.
Cisco is a culture that outsources a significant amount of work to its contract
manufacturers through its core and context framework. For instance, Cisco's core will
develop a new product and all the capabilities for testing the product. When the product
becomes more mature and all the bugs are fixed, Cisco will move the process to context
and deploy it to a contract manufacturer. This allows Cisco to move employees to the
next core process and continue innovating on future products. While this may sound
great on paper, core and context are continuously misused and confused across the
company. Core and context are part of Cisco's culture and used in everyday language for
every topic discussed; however, the definition of core and context differs for every
employee. As layers of core and context are revealed, the ownership of each process
becomes more ambiguous. Moving forward, Test Excellence will clearly define which
activities are core and which are context.
Different appearances also silo Cisco. While appearance is important, many types of
appearances are juxtaposed in the Cisco culture, which is evident when reviewing the
work process flows and the quality of the work deliverables. When walking around the
office space, desks, areas, and workspaces are "ad-hoc" and cluttered, analogous to the
way acquisitions were made, buildings built, and standardized documents created. The
office appearance is a stark contrast to the sleek appearance of many external Cisco
PowerPoint presentations and customer experience centers. In contrast, internal Cisco
documentation is cluttered and messy. For example, there are four different standardized
documents for test plans and employees suggest the current test strategy is "ad-hoc" in
nature. The strategy document is 223 pages long, cluttered, additive, and contains
strategies, tactics, methodologies and tools. So while buildings were built in what
appears to be an "ad-hoc" style across campus, the buildings were given the same look,
feel, and layout to represent a nice outside, public appearance.
132
While Cisco's culture is very open door, titles are very important and also increase the
value of appearance. Managers want to become directors, VPs, and SVPs. Recently a
new title was created, the executive vice president or EVP. In the old school way, people
were promoted by shipping a product to market rather than delaying the product due to
quality concerns. Today, a major culture shift is occurring within the company. The
culture is changing to a Six Sigma culture where the "diving catch" and "firefighting"
that were once considered core competencies are being replaced by a focus on worldclass quality.
Finally, Cisco is virtually silo'd by time zone, language, and the internet. It is almost
impossible for people to communicate efficiently across a global time zone. In fact, there
are manufacturing employees in San Jose, CA that work unreasonable hours such as 7am
- 2am to be able to directly talk with the appropriate Asian contract manufactures.
Employees are therefore allowed freedom to work from home, telecommute, and work at
a local Starbucks (or other hotspots) with wireless connectivity. Cisco encourages
employees to work by any means possible using Cisco on Cisco unified technologies. 44
In summary, "[f]rom a cultural perspective, organizations are social systems in which
people must work and live together, and therefore the management of meaning is as
critical as the management of money and production" (Carroll, 2006, p. 10). Cisco lives
by its culture, its espoused values, and its GEM tradition. Cisco is a culture of
innovation, having fun, and working hard. The culture has attracted talented people,
great intellectual property, and made Cisco successful. These values are preached and
managed to drive a one Cisco, one vision philosophy. Cisco's unique culture has been
very successful using its own technology to innovate and push the envelope of
technology. The company continues to do very well, focusing on acquiring new
technologies and talented people, while bringing innovative products to market. The Test
Excellence Initiative has bridged the structural, functional, and virtual silos that exist
44
Cisco on Cisco technology refers to utilizing any proprietary technology that allows employees to
communicate, collaborate, and work more efficiently and effectively
133
within the culture. In fact, the Component Quality Yield Model and the PCBA
PrioritizationAlgorithm are two collaborative tools based on cross company effort.
Using Six Sigma principals, both tools were created from best practices as defined by
employees from across the organization. As the culture adopts a Six Sigma philosophy,
these analytical tools should successfully be adopted within the organization as part of
the normal mode of operation.
5.5.
Combining the Three Lenses
Cisco is successful because of its culture, which evolved from the company's quick
success in the 90s. The silo'd structure, political, and cultural landscape emerged from
the enormous growth the company experienced. Based on the current three lens analysis,
there are various areas where Cisco can improve its organization. The Test Excellence
Initiative combined with the Component Quality Yield Optimization Model and the PCBA
PrioritizationAlgorithm are first steps on the journey to a Six Sigma culture. While the
strategic lens may show inefficiencies in how functional groups interact, the political and
cultural lens give insight into how there are many underlying networks and traditions that
connect Cisco together. To remain successful, Cisco needs to break down the barriers
that formed over time and learn how to effectively communicate across the company, not
just within each lens. Aligning the entire company to be quality focused is a step in the
right direction. In the future, Cisco's "Change Management Team" may benefit by using
the three lens tool to determine how the other lenses may be affected when new changes
are implemented.
Cisco is a very interesting company, composed of many smart, talented individuals
acquired over the last 20 years. Analyzing Cisco through the strategic, political, and
cultural lenses give a unique perspective into how the company really works and how the
company has evolved over time. The strategic, political, and cultural lenses are
intertwined. Each lens pulls and feeds off the other. Over time, each lens has been more
or less significant in defining what it is really like to work and get work accomplished at
Cisco. Every change at Cisco affects each lens. With change being constant, the lenses
are also constantly changing over time. The three lenses suggests that as Cisco shifts to a
134
Six Sigma quality oriented culture, the Component Quality Yield Model and PCBA
PriorityAlgorithm will be adopted if senior management continues to fully support the
initiative, political egos and agendas do not impede the progress, and yield and cost
improvements materialize.
5.6.
Summary
Despite all the complexities that lie within each lens analysis, Cisco has maintained a
significant competitive advantage and continues to be the market leader. Employees are
motivated and happy to work at Cisco while attrition rates continue to stay at the lowest
level compared to other Silicon Valley companies. 45 Its employees are determined to
keep making Cisco a better company through learning and continuous improvement. The
company can and will only get better. Thus, aspiring to be a world class manufacturing
organization and achieve a 60 yield process is very fitting for Cisco. When initially
discussing the ideas in this thesis with employees, many were very excited to participate,
stating that "change is exactly what Cisco needs."
Implementing the Component Quality Yield Model and the PCBA Prioritization
Algorithm will positively affect each organizational lens. Strategically, both models help
break down the silos within the culture, using best practices and opening up new channels
of communication. Politically, both models help move the power from the top-down
decision making method to the highly collaborative method, where teams make better,
more informed decisions based on the data. Culturally, both models aim to achieve a 6a
yield process which lends itself directly to the Six Sigma culture the CEO is trying to
establish in the greater Cisco organization. Overall, the organization is ready to support
change and the organizational design analysis justifies the right time for change is now!
45 Chambers, John. Live Presentation. CEO of Cisco Systems, Inc. 2007
135
This page has been intentionally left blank
136
6. Chapter 6: Conclusion
Cisco is at an inflection point in both its manufacturing capabilities and its organizational design.
To become a world class PCBA manufacturer, the firm aspires to achieve a 6a yield process.
Doing so requires a change in the organization's culture to enable the correct behaviors to
become world class.
This thesis analyzed three points that will help Cisco along its journey to Six Sigma:
Hypothesis 1: The Component Ouality Yield Model
Yields and costs can be optimized for an entire portfolio of PCBA products by selecting the
appropriate components based on component quality and cost specifications. Understanding the
implications of component selection will simultaneously improve yields at lower costs. The
Component Quality Yield Optimization Model is a tool that will allow design engineers,
component sourcing engineers, and manufacturing managers to make better, more informed
business decisions and help Cisco take one step closer to achieving a 6a yield process on every
single PCBA manufactured.
Hypothesis 2: The PCBA Prioritization Algorithm
Limited manufacturing resources can be allocated more efficiently to resolve yield issues when
prioritizing PCBAs holistically, using the same methodology across the portfolio of products
rather than locally, where prioritization differs by each manufacturing site. Knowing where the
most mission critical problems are amongst the thousands of PCBAs manufactured allows Cisco
to solve the most important problems first. By prioritizing manufacturing resources more
effectively, Cisco will realize increased yield improvements and lower costs quicker. Thus, the
PCBA PrioritizationAlgorithm offers another tool that will aid Cisco in the quest to achieve a 6a
yield process on every single PCBA manufactured.
137
Organizational Design Analysis:
The last chapter of this thesis reviewed the importance of understanding the current
organizational design in order to foster an environment ready for change, which will enable
Cisco to be successful in becoming a world class PCBA manufacturer. In order to achieve a 6a
yield process on every PCBA, the company must have an organization willing to make the
journey. Understanding the current landscape of the organization and how to effectively drive
change will aid in Cisco's success.
The results from the two hypotheses and the organizational design analysis are only steps in the
journey towards Six Sigma. This thesis provides analytical tools that can process current
manufacturing data to develop an optimal solution. By utilizing the Component Quality Yield
Optimization Model and the PCBA PrioritizationAlgorithm, Cisco will be able to better improve
and control PCBA yields and costs. With higher yields and lower costs, test plans can then be
optimized better over the lifecycle of a product. Thus, investments in the tools and the work to
implement the solutions will lead to improved capabilities that will drive further yield and cost
improvements. In turn, this will allow less capital to be spent on future tests. The process
becomes a positive reinforcing loop (Repenning & Sterman, 2001). Realizing that change will
not happen overnight, Cisco is prepared to develop new capabilities and ensure the Test
Excellence initiative is successful. The program will need key internal leadership and drive to
keep the cross company teams fully engaged and aligned to the 60 goals. Further yield and cost
improvement projects will be identified and implemented, and step by step, Cisco will be closer
to its goal of becoming a world class, Six Sigma manufacturer.
138
Bibliography
[ 1]
Ahn, BS, & Choi, SH. (2007). ERP system selection using a simulation-based AHP
approach: a case of a Korean homeshopping company. Journalof the Operations
Research Society. January2007, Volume 59, p 3 2 2 - 330.
[2 ]
Arrufiada, Benito & Vazquez, Xos6 H. (2006). When Your Contract Manufacturer
Becomes Your Competitor. HarvardBusiness Review. September 2006,p1 3 5 - 144.
[3 ]
Ashmore, Clive. (2000). From SPC to DPMO. CircuitsAssembly. March 2000. Vol. 17
Issue, p 2 5.
[4 ]
Bafiuelas, Ricardo& Antony, Jiju. (2005). Six Sigma or design for Six Sigma? The TQM
Magazine. 2005. Vol. 16. Issue 4, p 2 5 0 -2 6 3 .
[5 ]
Barney, Matt & McCarty, Tom. (2003). The New 60: A Leader's Guide to Achieving
Rapid Business Improvement and Sustainable Results. Upper Saddle River, New
Jersey: Prentis Hall PTR.
[6 ]
Bigelow, Peter. (2006). Stuff Happens. PrintedCircuit Design & Manufacture, Oct2006,
Vol. 23 Issue 10, p14.
[7]
Brealey, Richard A., Myers, Stewart C., & Allen, Franklin. (2006). Principalsof
CorporateFinance (8th ed.). Boston: McGraw-Hill Irwin.
[ 8]
Britton, Keith J. & Schaible, Dawn M. (2003). Testing in NASA Human-Rated
Spacecraft Programs: How much is just enough? SM Thesis, MassachusettsInstitute
of Technology, June 2003.
[ 9]
Brooke, Lindsay. (2005). Denso drives ahead: A plan to double global sales by 2005 has
Japan'ssupplier giant thinking differently. Retrieved April 30, 2008, from
http://findarticles.com/p/articles/mi_m3012/is_2_182/ai_84377695
[ 10 ] Bruns, Adam. (2003). MicroprocessorMakers Seek Solid State: Even with overcapacity,
some semiconductorfirms are stealinga march on the future. Retrieved April 30,
2008, from http://www.siteselection.com/issues/2003/sep/p565/
[ 11 ] Carroll, John S. (2006). Introduction to Organizational Analysis: The Three Lenses. MIT
Sloan School, Revised, June 2006
[ 12 ] Chen, Jason T. (2003). A Process for Improving Early Life Failure Response. SM Thesis,
MassachusettsInstitute of Technology, June 2003.
139
[ 13 ] Collins, Jim.(2001) Good to Great: Why Some Companies Make the Leap ... and Others'
Don't. New York: HarperCollins Publishers Inc.
[ 14 ] Ellis, Kimberley P., Vittes, Fernando J., & Kobza, John E. (2001) Optimizing the
Performance of a Surface Mount Placement Machine. IEEE Transactionson
Electronics PackagingManufacturing,July 2001. Volume 24, number 3, p1 6 0 - 1 70.
[ 15 ] Franklin, Joel P., & Niemeier, Debbie A. (1998). The Prioritization of Mobility
Improvements Using a Multicriteria Prioritization Algorithm. Department of Civil
and EnvironmentalEngineering& Institute of TransportationStudies, University of
California.July 1998.
[ 16 ] Freidman, Thomas L. (2006) The World is Flat.A BriefHistory of the Twenty-first
Century. New York: Farrar, Straus and Giroux.
[ 17 ] Fine, Charles H. (1998). Clockspeed: Winning Industry Control in the Age of Temporary
Advantage. Reading, Massachusetts: Perseus Books.
[ 18 ] Fine, Charles H. (2008). MIT Lecture Notes. Spring 2008. Course 15.769: Operations
Strategy
[ 19 ] Gamarnik, David. (2006). MITLecture Notes. Summer 2006. Course 15.066: Systems
Optimization andAnalysis for Manufacturing.
[ 20 ] Gilbert, James M., Bell, Ian M., & Johnson, David R. (2005). Circuit Design
Optimization Based on Quality Cost Estimation. Quality and Reliability Engineering
International.January2005. Volume 21, p 3 6 7 - 386.
[ 21 ] Go, Julie W. (2007). Case Studies in DSM: Utilizing the Design Structure Matrix to
Improve New Product Introduction. SM Thesis, Massachusetts Institute of
Technology, June 2007.
[ 22 ] Goldratt, Eliyahu M. & Cox, Jeff. (1992). The Goal: A Processof OngoingImprovement
(Second Revised Edition). Great Barrington, Massachusetts: North River Press.
[ 23 ] Goodman, Arnold; Kamath, Chandrika; & Kumar, Vipin. (2008). Data Analysis in the
21st Century. StatisticalAnalysis andDataMining. Volume 1 Issue 1. February
2008.
[ 24 ] Haag, Lance Edward. (1995). Effective Use of Test Data for Quality Improvement and
Cycle Time Reduction in Radio System Manufacturing. SM Thesis, Massachusetts
Institute of Technology, June 1995.
[ 25 ] Hammer, Michael. (2004). Deep Change: How Operational Innovation can Transform
your Company. HarvardBusiness Review, April 2004, p85 - 93.
140
[ 26 ] Hany et al. (1992). Six Sigma Productibility Analysis and Process Characterization.
1992, pp. 1-1 to 1-5, 2-4-2 and 5-11 to 5-13.
[ 27 ] Hebalkar, Tejaswini. (2007). Re-Architecting the Failure Analysis Supply Chain. SM
Thesis, MassachusettsInstitute of Technology, June 2007.
[ 28 ] Ho, W. & Ji, P. (2005). PCB assembly line assignment: a genetic algorithm
approach. Journalof Manufacturing Technology Management, 16(5/6), p 6 8 2 -6 9 2 .
[ 29 ] Jaimes, Estaban A Guerrero. (2003). Collaborative Design for Supply Chain: Including
Strategic and Tactical Supply-Chain Considerations in Product Design &
Development. SM Thesis, MassachusettsInstitute of Technology, May 2003.
[ 30 ] Kellogg, Kate. (2006). MITLecture Notes. Fall2006. Course 15.277: Organizational
Processes.
[ 31 ] Klein, Janice A. (2004). True Change: How Outsiders on the Inside Get Things Done in
Organizations.San Francisco: Jossey-Bass.
[ 32 ] Knudsen, David Charles.(2001). Application of Statistical Quality Control to Improve
Yields and Rationalize Testing in a Low Volume Manufacturing Facility. SM Thesis,
MassachusettsInstitute of Technology, June 2001.
[ 33 ] Kwak, Young Hoon & Anbari, Frank T. (2006). Benefits, obstacles, and future of six
sigma approach. TechnovationVolume 26, Issues 5-6, May-June 2006, p 70 8 - 71 5 .
[ 34 ] Markoff, John. (1990). Five U.S Chip Makers to Work with Toyota. Retreived August
30, 2008 from
http://query.nytimes.com/gst/fullpage.html?res=9COCE2DC 1E31F93AA15755COA9
66958260.
[ 35 ] Michel, C.; & Reinosa, R.D. (2004). Test strategy cost model innovations. Test
Conference, 2004. Proceedings.ITC 2004. International 26-28 Oct. 2004, p384392.
[ 36 ] Myers, M. Parker. (1996). Improving Production Testing of RF Products in a Noisy
Measurement Environment. SM Thesis, MassachusettsInstitute of Technology, June
1996.
[ 37 ] Nelson, Erik Tighe. (2000). Optimizing Production Testing in the Electronics
Manufacturing Industry. SM Thesis, MassachusettsInstitute of Technology, June
2000.
[ 38 ] Nystedt, Dan. (2008). Flash price drop spurring innovation. IDG News Services.
February2008,pl.
141
[ 39 ] Ohno, Taiichi. (1988). Toyota ProductionSystem: Beyond Large-Scale Production.New
York: Productivity Press.
[ 40 ] Olson, D.L. (2004). Comparison of weights in TOPSIS models. Mathematicaland
Computer Modeling. Volume 40, Issues 7-8. October 2004, p 72 1-72 7 .
[ 41 ] O'Neil, William. (2004). The Successful Investor. New York: McGraw-Hill.
[ 42 ] Oresjo, Stig. (2003). Selecting the Optimal Test Strategy. CircuitsAssembly. July 2003,
p14-16.
[ 43 ] Oresjo, Stig. (2005). Step by Step Test and Inspection. SMT. October 2005, Volume 19,
Issue 10, p 4 4 -5 0 .
[ 44 ] Parker, Kevin. (2008) What does it mean to say it's model-based? Manufacturing
Business Technology, Mar2008, Vol. 26 Issue 3, p 2 .
[ 45 ] Peterson, Michael L. (1994). Testing for Time-Dependent Failure in Electronic Products.
SM Thesis, MassachusettsInstitute of Technology, May 1994.
[ 46 ] Porter, Michael E.. (1996). Operational Effectiveness is Not Strategy. HarvardBusiness
Review, November - December 1996.p61 - 78.
[ 47 ] Quinn, Dan. (2005). Six Sigma Leaders. Quality, 44(3), p 8 0 .
[ 48 ] Rae, Alan; Pfahl, Robert C. & Richardson, Charles. (2008). Research Priorities for the
Electronics Industry. PrintedCircuitDesign & Manufacture,Jan2008, Vol. 25 Issue
1, p 2 7 -2 9 .
[49]
Ragsdale, Cliff T. (2004). Spreadsheet Modeling & Decision Analysis: A Practical
Introductionto Management Science (4th ed.). United States: Thomson South-
Western.
[ 50 ] Repenning, Nelson P. & Sterman, John D. (2001). Nobody Ever Gets Credit for Fixing
Problems that Never Happen: Creating and Sustaining Process Improvement.
CaliforniaManagementReview. Vol. 43, No 4. Summer 2001. p 6 4 -8 8 .
[ 51 ] Shanian, A. & Savadogo, O. (2006). TOPSIS multiple-criteria decision support analysis
for material selection of metallic bipolar plates for polymer electrolyte fuel cell.
Journalof Power Sources. Volume 159, Issue 2, 22 September 2006,p1095-1104.
[ 52 ] Skinner, Wickham. (1969). Manufacturing - missing link in corporate strategy. Harvard
Business Review, May - June 1969, p1 3 6 - 145.
142
[ 53 ] Teixeira de Almeida, Adiel. (2007). Mulitcriteria decision model for outsourcing
contracts selection based on utility function and ELECTRE method. Computers and
OperationsResearch. Volume 34, Issue 12, December 2007,p3569 - 3574.
[ 54 ] Tien-Chin Wang & Tsung-Han Chang. (2007). Application of TOPSIS in evaluating
initial training aircraft under a fuzzy environment. Expert Systems with Applications.
Volume 33, Issue 4, November 2007,p870 - 880.
[ 55 ] Tong, J.P.C; Tsung, F.; & Yen, B.P.C. (2004). A DMAIC approach to printed circuit
board quality improvement. The InternationalJournalofAdvance Manufacturing.
Publishedby Springer London. Volume 23, Numbers 7-8. April 2004, pl - 9.
[ 56 ] Utterback, J.M. (1994). Mastering the Dynamics of Innovation. HarvardBusiness School
Press, Boston Massachusetts.
[ 57 ] Van Hop, Nguyen, & Tabucanon, M.T. (2001). Set-up problem for a line of machines in
a PCB assembly process planning: a multiple criteria decision making process.
InternaltionalJournalof ComputerIntegratedManufacturing. August 2001, Volume
14, Issue 4, p 3 4 3 - 352.
[ 58 ] Vlasak, Dean Allen. (1990). Process Verification Test Cell for Low-Volume, ASIC-like
MMIC Production Environment. SM Thesis, MassachusettsInstitute of Technology,
August 1990.
[ 59 ] Wang, Paul P.E.; Martinez-Vargas, Jorge; Gill, Dinesh; Iyer, Ramamoorthy Ganapthy; &
Kauss, Dan. (2004). Total Process Control and Management, Part II. Circuits
Assembly. May 2004. Vol. 15 Issue 5, p 2 2 -2 7 .
[ 60 ] Womack, James P., Jones, Daniel T., & Roos, Daniel. (1990). The Machine that Changed
the World. New York: Free Press.
[ 61 ] Womack, James P., & Jones, Daniel T. (1996). Lean Thinking: Banish Waste and Create
Wealth in Your Corporation.New York: Simon & Schuster.
[ 62 ] Yoon, K. & Hwang, C.L. (1980). Multiple Attribute Decision Making Methods and
Applications. A State of the Art Survey, Springer Verlag, Berlin, 1980.
[ 63 ] Yoon, K. (1980). System selection by multiple attribute decision making. Ph.D.
Dissertation,KansasState University, Manhattan,Kansas, 1980.
143